20220617.md

ArXiv cs.CV --Fri, 17 Jun 2022

1.Unbiased 4D: Monocular 4D Reconstruction with a Neural Deformation Model ⬇️

Capturing general deforming scenes is crucial for many computer graphics and vision applications, and it is especially challenging when only a monocular RGB video of the scene is available. Competing methods assume dense point tracks, 3D templates, large-scale training datasets, or only capture small-scale deformations. In contrast to those, our method, Ub4D, makes none of these assumptions while outperforming the previous state of the art in challenging scenarios. Our technique includes two new, in the context of non-rigid 3D reconstruction, components, i.e., 1) A coordinate-based and implicit neural representation for non-rigid scenes, which enables an unbiased reconstruction of dynamic scenes, and 2) A novel dynamic scene flow loss, which enables the reconstruction of larger deformations. Results on our new dataset, which will be made publicly available, demonstrate the clear improvement over the state of the art in terms of surface reconstruction accuracy and robustness to large deformations. Visit the project page this https URL.

2.SHIFT: A Synthetic Driving Dataset for Continuous Multi-Task Domain Adaptation ⬇️

Adapting to a continuously evolving environment is a safety-critical challenge inevitably faced by all autonomous driving systems. Existing image and video driving datasets, however, fall short of capturing the mutable nature of the real world. In this paper, we introduce the largest multi-task synthetic dataset for autonomous driving, SHIFT. It presents discrete and continuous shifts in cloudiness, rain and fog intensity, time of day, and vehicle and pedestrian density. Featuring a comprehensive sensor suite and annotations for several mainstream perception tasks, SHIFT allows investigating the degradation of a perception system performance at increasing levels of domain shift, fostering the development of continuous adaptation strategies to mitigate this problem and assess model robustness and generality. Our dataset and benchmark toolkit are publicly available at www.vis.xyz/shift.

3.Virtual Correspondence: Humans as a Cue for Extreme-View Geometry ⬇️

Recovering the spatial layout of the cameras and the geometry of the scene from extreme-view images is a longstanding challenge in computer vision. Prevailing 3D reconstruction algorithms often adopt the image matching paradigm and presume that a portion of the scene is co-visible across images, yielding poor performance when there is little overlap among inputs. In contrast, humans can associate visible parts in one image to the corresponding invisible components in another image via prior knowledge of the shapes. Inspired by this fact, we present a novel concept called virtual correspondences (VCs). VCs are a pair of pixels from two images whose camera rays intersect in 3D. Similar to classic correspondences, VCs conform with epipolar geometry; unlike classic correspondences, VCs do not need to be co-visible across views. Therefore VCs can be established and exploited even if images do not overlap. We introduce a method to find virtual correspondences based on humans in the scene. We showcase how VCs can be seamlessly integrated with classic bundle adjustment to recover camera poses across extreme views. Experiments show that our method significantly outperforms state-of-the-art camera pose estimation methods in challenging scenarios and is comparable in the traditional densely captured setup. Our approach also unleashes the potential of multiple downstream tasks such as scene reconstruction from multi-view stereo and novel view synthesis in extreme-view scenarios.

4.Unified Fourier-based Kernel and Nonlinearity Design for Equivariant Networks on Homogeneous Spaces ⬇️

We introduce a unified framework for group equivariant networks on homogeneous spaces derived from a Fourier perspective. We address the case of feature fields being tensor valued before and after a convolutional layer. We present a unified derivation of kernels via the Fourier domain by taking advantage of the sparsity of Fourier coefficients of the lifted feature fields. The sparsity emerges when the stabilizer subgroup of the homogeneous space is a compact Lie group. We further introduce an activation method via an elementwise nonlinearity on the regular representation after lifting and projecting back to the field through an equivariant convolution. We show that other methods treating features as the Fourier coefficients in the stabilizer subgroup are special cases of our activation. Experiments on $SO(3)$ and $SE(3)$ show state-of-the-art performance in spherical vector field regression, point cloud classification, and molecular completion.

5.Controllable 3D Face Synthesis with Conditional Generative Occupancy Fields ⬇️

Capitalizing on the recent advances in image generation models, existing controllable face image synthesis methods are able to generate high-fidelity images with some levels of controllability, e.g., controlling the shapes, expressions, textures, and poses of the generated face images. However, these methods focus on 2D image generative models, which are prone to producing inconsistent face images under large expression and pose changes. In this paper, we propose a new NeRF-based conditional 3D face synthesis framework, which enables 3D controllability over the generated face images by imposing explicit 3D conditions from 3D face priors. At its core is a conditional Generative Occupancy Field (cGOF) that effectively enforces the shape of the generated face to commit to a given 3D Morphable Model (3DMM) mesh. To achieve accurate control over fine-grained 3D face shapes of the synthesized image, we additionally incorporate a 3D landmark loss as well as a volume warping loss into our synthesis algorithm. Experiments validate the effectiveness of the proposed method, which is able to generate high-fidelity face images and shows more precise 3D controllability than state-of-the-art 2D-based controllable face synthesis methods. Find code and demo at this https URL.

6.MixGen: A New Multi-Modal Data Augmentation ⬇️

Data augmentation is a necessity to enhance data efficiency in deep learning. For vision-language pre-training, data is only augmented either for images or for text in previous works. In this paper, we present MixGen: a joint data augmentation for vision-language representation learning to further improve data efficiency. It generates new image-text pairs with semantic relationships preserved by interpolating images and concatenating text. It's simple, and can be plug-and-played into existing pipelines. We evaluate MixGen on four architectures, including CLIP, ViLT, ALBEF and TCL, across five downstream vision-language tasks to show its versatility and effectiveness. For example, adding MixGen in ALBEF pre-training leads to absolute performance improvements on downstream tasks: image-text retrieval (+6.2% on COCO fine-tuned and +5.3% on Flicker30K zero-shot), visual grounding (+0.9% on RefCOCO+), visual reasoning (+0.9% on NLVR$^{2}$), visual question answering (+0.3% on VQA2.0), and visual entailment (+0.4% on SNLI-VE).

7.Spatially-Adaptive Multilayer Selection for GAN Inversion and Editing ⬇️

Existing GAN inversion and editing methods work well for aligned objects with a clean background, such as portraits and animal faces, but often struggle for more difficult categories with complex scene layouts and object occlusions, such as cars, animals, and outdoor images. We propose a new method to invert and edit such complex images in the latent space of GANs, such as StyleGAN2. Our key idea is to explore inversion with a collection of layers, spatially adapting the inversion process to the difficulty of the image. We learn to predict the "invertibility" of different image segments and project each segment into a latent layer. Easier regions can be inverted into an earlier layer in the generator's latent space, while more challenging regions can be inverted into a later feature space. Experiments show that our method obtains better inversion results compared to the recent approaches on complex categories, while maintaining downstream editability. Please refer to our project page at this https URL.

8.OmniMAE: Single Model Masked Pretraining on Images and Videos ⬇️

Transformer-based architectures have become competitive across a variety of visual domains, most notably images and videos. While prior work has studied these modalities in isolation, having a common architecture suggests that one can train a single unified model for multiple visual modalities. Prior attempts at unified modeling typically use architectures tailored for vision tasks, or obtain worse performance compared to single modality models. In this work, we show that masked autoencoding can be used to train a simple Vision Transformer on images and videos, without requiring any labeled data. This single model learns visual representations that are comparable to or better than single-modality representations on both image and video benchmarks, while using a much simpler architecture. In particular, our single pretrained model can be finetuned to achieve 86.5% on ImageNet and 75.3% on the challenging Something Something-v2 video benchmark. Furthermore, this model can be learned by dropping 90% of the image and 95% of the video patches, enabling extremely fast training.

9.FWD: Real-time Novel View Synthesis with Forward Warping and Depth ⬇️

Novel view synthesis (NVS) is a challenging task requiring systems to generate photorealistic images of scenes from new viewpoints, where both quality and speed are important for applications. Previous image-based rendering (IBR) methods are fast, but have poor quality when input views are sparse. Recent Neural Radiance Fields (NeRF) and generalizable variants give impressive results but are not real-time. In our paper, we propose a generalizable NVS method with sparse inputs, called FWD, which gives high-quality synthesis in real-time. With explicit depth and differentiable rendering, it achieves competitive results to the SOTA methods with 130-1000x speedup and better perceptual quality. If available, we can seamlessly integrate sensor depth during either training or inference to improve image quality while retaining real-time speed. With the growing prevalence of depths sensors, we hope that methods making use of depth will become increasingly useful.

10.Beyond Supervised vs. Unsupervised: Representative Benchmarking and Analysis of Image Representation Learning ⬇️

By leveraging contrastive learning, clustering, and other pretext tasks, unsupervised methods for learning image representations have reached impressive results on standard benchmarks. The result has been a crowded field - many methods with substantially different implementations yield results that seem nearly identical on popular benchmarks, such as linear evaluation on ImageNet. However, a single result does not tell the whole story. In this paper, we compare methods using performance-based benchmarks such as linear evaluation, nearest neighbor classification, and clustering for several different datasets, demonstrating the lack of a clear front-runner within the current state-of-the-art. In contrast to prior work that performs only supervised vs. unsupervised comparison, we compare several different unsupervised methods against each other. To enrich this comparison, we analyze embeddings with measurements such as uniformity, tolerance, and centered kernel alignment (CKA), and propose two new metrics of our own: nearest neighbor graph similarity and linear prediction overlap. We reveal through our analysis that in isolation, single popular methods should not be treated as though they represent the field as a whole, and that future work ought to consider how to leverage the complimentary nature of these methods. We also leverage CKA to provide a framework to robustly quantify augmentation invariance, and provide a reminder that certain types of invariance will be undesirable for downstream tasks.

11.Real-World Single Image Super-Resolution Under Rainy Condition ⬇️

Image super-resolution is an important research area in computer vision that has a wide variety of applications including surveillance, medical imaging etc. Real-world signal image super-resolution has become very popular now-a-days due to its real-time application. There are still a lot of scopes to improve real-world single image super-resolution specially during challenging weather scenarios. In this paper, we have proposed a new algorithm to perform real-world single image super-resolution during rainy condition. Our proposed method can mitigate the influence of rainy conditions during image super-resolution. Our experiment results show that our proposed algorithm can perform image super-resolution decreasing the negative effects of the rain.

12.Realistic One-shot Mesh-based Head Avatars ⬇️

We present a system for realistic one-shot mesh-based human head avatars creation, ROME for short. Using a single photograph, our model estimates a person-specific head mesh and the associated neural texture, which encodes both local photometric and geometric details. The resulting avatars are rigged and can be rendered using a neural network, which is trained alongside the mesh and texture estimators on a dataset of in-the-wild videos. In the experiments, we observe that our system performs competitively both in terms of head geometry recovery and the quality of renders, especially for the cross-person reenactment. See results this https URL

13.iBoot: Image-bootstrapped Self-Supervised Video Representation Learning ⬇️

Learning visual representations through self-supervision is an extremely challenging task as the network needs to sieve relevant patterns from spurious distractors without the active guidance provided by supervision. This is achieved through heavy data augmentation, large-scale datasets and prohibitive amounts of compute. Video self-supervised learning (SSL) suffers from added challenges: video datasets are typically not as large as image datasets, compute is an order of magnitude larger, and the amount of spurious patterns the optimizer has to sieve through is multiplied several fold. Thus, directly learning self-supervised representations from video data might result in sub-optimal performance. To address this, we propose to utilize a strong image-based model, pre-trained with self- or language supervision, in a video representation learning framework, enabling the model to learn strong spatial and temporal information without relying on the video labeled data. To this end, we modify the typical video-based SSL design and objective to encourage the video encoder to \textit{subsume} the semantic content of an image-based model trained on a general domain. The proposed algorithm is shown to learn much more efficiently (i.e. in less epochs and with a smaller batch) and results in a new state-of-the-art performance on standard downstream tasks among single-modality SSL methods.

14.Adversarial Patch Attacks and Defences in Vision-Based Tasks: A Survey ⬇️

Adversarial attacks in deep learning models, especially for safety-critical systems, are gaining more and more attention in recent years, due to the lack of trust in the security and robustness of AI models. Yet the more primitive adversarial attacks might be physically infeasible or require some resources that are hard to access like the training data, which motivated the emergence of patch attacks. In this survey, we provide a comprehensive overview to cover existing techniques of adversarial patch attacks, aiming to help interested researchers quickly catch up with the progress in this field. We also discuss existing techniques for developing detection and defences against adversarial patches, aiming to help the community better understand this field and its applications in the real world.

15.Rank the triplets: A ranking-based multiple instance learning framework for detecting HPV infection in head and neck cancers using routine H&E images ⬇️

The aetiology of head and neck squamous cell carcinoma (HNSCC) involves multiple carcinogens such as alcohol, tobacco and infection with human papillomavirus (HPV). As the HPV infection influences the prognosis, treatment and survival of patients with HNSCC, it is important to determine the HPV status of these tumours. In this paper, we propose a novel triplet-ranking loss function and a multiple instance learning pipeline for HPV status prediction. This achieves a new state-of-the-art performance in HPV detection using only the routine H&E stained WSIs on two HNSCC cohorts. Furthermore, a comprehensive tumour microenvironment profiling was performed, which characterised the unique patterns between HPV+/- HNSCC from genomic, immunology and cellular perspectives. Positive correlations of the proposed score with different subtypes of T cells (e.g. T cells follicular helper, CD8+ T cells), and negative correlations with macrophages and connective cells (e.g. fibroblast) were identified, which is in line with clinical findings. Unique gene expression profiles were also identified with respect to HPV infection status, and is in line with existing findings.

16.Simple and Efficient Architectures for Semantic Segmentation ⬇️

Though the state-of-the architectures for semantic segmentation, such as HRNet, demonstrate impressive accuracy, the complexity arising from their salient design choices hinders a range of model acceleration tools, and further they make use of operations that are inefficient on current hardware. This paper demonstrates that a simple encoder-decoder architecture with a ResNet-like backbone and a small multi-scale head, performs on-par or better than complex semantic segmentation architectures such as HRNet, FANet and DDRNets. Naively applying deep backbones designed for Image Classification to the task of Semantic Segmentation leads to sub-par results, owing to a much smaller effective receptive field of these backbones. Implicit among the various design choices put forth in works like HRNet, DDRNet, and FANet are networks with a large effective receptive field. It is natural to ask if a simple encoder-decoder architecture would compare favorably if comprised of backbones that have a larger effective receptive field, though without the use of inefficient operations like dilated convolutions. We show that with minor and inexpensive modifications to ResNets, enlarging the receptive field, very simple and competitive baselines can be created for Semantic Segmentation. We present a family of such simple architectures for desktop as well as mobile targets, which match or exceed the performance of complex models on the Cityscapes dataset. We hope that our work provides simple yet effective baselines for practitioners to develop efficient semantic segmentation models.

17.Open-Set Recognition with Gradient-Based Representations ⬇️

Neural networks for image classification tasks assume that any given image during inference belongs to one of the training classes. This closed-set assumption is challenged in real-world applications where models may encounter inputs of unknown classes. Open-set recognition aims to solve this problem by rejecting unknown classes while classifying known classes correctly. In this paper, we propose to utilize gradient-based representations obtained from a known classifier to train an unknown detector with instances of known classes only. Gradients correspond to the amount of model updates required to properly represent a given sample, which we exploit to understand the model's capability to characterize inputs with its learned features. Our approach can be utilized with any classifier trained in a supervised manner on known classes without the need to model the distribution of unknown samples explicitly. We show that our gradient-based approach outperforms state-of-the-art methods by up to 11.6% in open-set classification.

18.Delving into the Scale Variance Problem in Object Detection ⬇️

Object detection has made substantial progress in the last decade, due to the capability of convolution in extracting local context of objects. However, the scales of objects are diverse and current convolution can only process single-scale input. The capability of traditional convolution with a fixed receptive field in dealing with such a scale variance problem, is thus limited. Multi-scale feature representation has been proven to be an effective way to mitigate the scale variance problem. Recent researches mainly adopt partial connection with certain scales, or aggregate features from all scales and focus on the global information across the scales. However, the information across spatial and depth dimensions is ignored. Inspired by this, we propose the multi-scale convolution (MSConv) to handle this problem. Taking into consideration scale, spatial and depth information at the same time, MSConv is able to process multi-scale input more comprehensively. MSConv is effective and computationally efficient, with only a small increase of computational cost. For most of the single-stage object detectors, replacing the traditional convolutions with MSConvs in the detection head can bring more than 2.5% improvement in AP (on COCO 2017 dataset), with only 3% increase of FLOPs. MSConv is also flexible and effective for two-stage object detectors. When extended to the mainstream two-stage object detectors, MSConv can bring up to 3.0% improvement in AP. Our best model under single-scale testing achieves 48.9% AP on COCO 2017 \textit{test-dev} split, which surpasses many state-of-the-art methods.

19.Multi scale Feature Extraction and Fusion for Online Knowledge Distillation ⬇️

Online knowledge distillation conducts knowledge transfer among all student models to alleviate the reliance on pre-trained models. However, existing online methods rely heavily on the prediction distributions and neglect the further exploration of the representational knowledge. In this paper, we propose a novel Multi-scale Feature Extraction and Fusion method (MFEF) for online knowledge distillation, which comprises three key components: Multi-scale Feature Extraction, Dual-attention and Feature Fusion, towards generating more informative feature maps for distillation. The multiscale feature extraction exploiting divide-and-concatenate in channel dimension is proposed to improve the multi-scale representation ability of feature maps. To obtain more accurate information, we design a dual-attention to strengthen the important channel and spatial regions adaptively. Moreover, we aggregate and fuse the former processed feature maps via feature fusion to assist the training of student models. Extensive experiments on CIF AR-10, CIF AR-100, and CINIC-10 show that MFEF transfers more beneficial representational knowledge for distillation and outperforms alternative methods among various network architectures

20.Adapting Self-Supervised Vision Transformers by Probing Attention-Conditioned Masking Consistency ⬇️

Visual domain adaptation (DA) seeks to transfer trained models to unseen, unlabeled domains across distribution shift, but approaches typically focus on adapting convolutional neural network architectures initialized with supervised ImageNet representations. In this work, we shift focus to adapting modern architectures for object recognition -- the increasingly popular Vision Transformer (ViT) -- and modern pretraining based on self-supervised learning (SSL). Inspired by the design of recent SSL approaches based on learning from partial image inputs generated via masking or cropping -- either by learning to predict the missing pixels, or learning representational invariances to such augmentations -- we propose PACMAC, a simple two-stage adaptation algorithm for self-supervised ViTs. PACMAC first performs in-domain SSL on pooled source and target data to learn task-discriminative features, and then probes the model's predictive consistency across a set of partial target inputs generated via a novel attention-conditioned masking strategy, to identify reliable candidates for self-training. Our simple approach leads to consistent performance gains over competing methods that use ViTs and self-supervised initializations on standard object recognition benchmarks. Code available at this https URL

21.HaGRID - HAnd Gesture Recognition Image Dataset ⬇️

In this paper, we introduce an enormous dataset HaGRID (HAnd Gesture Recognition Image Dataset) for hand gesture recognition (HGR) systems. This dataset contains 552,992 samples divided into 18 classes of gestures. The annotations consist of bounding boxes of hands with gesture labels and markups of leading hands. The proposed dataset allows for building HGR systems, which can be used in video conferencing services, home automation systems, the automotive sector, services for people with speech and hearing impairments, etc. We are especially focused on interaction with devices to manage them. That is why all 18 chosen gestures are functional, familiar to the majority of people, and may be an incentive to take some action. In addition, we used crowdsourcing platforms to collect the dataset and took into account various parameters to ensure data diversity. We describe the challenges of using existing HGR datasets for our task and provide a detailed overview of them. Furthermore, the baselines for the hand detection and gesture classification tasks are proposed.

22.Selective Multi-Scale Learning for Object Detection ⬇️

Pyramidal networks are standard methods for multi-scale object detection. Current researches on feature pyramid networks usually adopt layer connections to collect features from certain levels of the feature hierarchy, and do not consider the significant differences among them. We propose a better architecture of feature pyramid networks, named selective multi-scale learning (SMSL), to address this issue. SMSL is efficient and general, which can be integrated in both single-stage and two-stage detectors to boost detection performance, with nearly no extra inference cost. RetinaNet combined with SMSL obtains 1.8% improvement in AP (from 39.1% to 40.9%) on COCO dataset. When integrated with SMSL, two-stage detectors can get around 1.0% improvement in AP.

23.Online Segmentation of LiDAR Sequences: Dataset and Algorithm ⬇️

Roof-mounted spinning LiDAR sensors are widely used by autonomous vehicles, driving the need for real-time processing of 3D point sequences. However, most LiDAR semantic segmentation datasets and algorithms split these acquisitions into $360^\circ$ frames, leading to acquisition latency that is incompatible with realistic real-time applications and evaluations. We address this issue with two key contributions. First, we introduce HelixNet, a $10$ billion point dataset with fine-grained labels, timestamps, and sensor rotation information that allows an accurate assessment of real-time readiness of segmentation algorithms. Second, we propose Helix4D, a compact and efficient spatio-temporal transformer architecture specifically designed for rotating LiDAR point sequences. Helix4D operates on acquisition slices that correspond to a fraction of a full rotation of the sensor, significantly reducing the total latency. We present an extensive benchmark of the performance and real-time readiness of several state-of-the-art models on HelixNet and SemanticKITTI. Helix4D reaches accuracy on par with the best segmentation algorithms with a reduction of more than $5\times$ in terms of latency and $50\times$ in model size. Code and data are available at: this https URL

24.Asymptotic Soft Cluster Pruning for Deep Neural Networks ⬇️

Filter pruning method introduces structural sparsity by removing selected filters and is thus particularly effective for reducing complexity. Previous works empirically prune networks from the point of view that filter with smaller norm contributes less to the final results. However, such criteria has been proven sensitive to the distribution of filters, and the accuracy may hard to recover since the capacity gap is fixed once pruned. In this paper, we propose a novel filter pruning method called Asymptotic Soft Cluster Pruning (ASCP), to identify the redundancy of network based on the similarity of filters. Each filter from over-parameterized network is first distinguished by clustering, and then reconstructed to manually introduce redundancy into it. Several guidelines of clustering are proposed to better preserve feature extraction ability. After reconstruction, filters are allowed to be updated to eliminate the effect caused by mistakenly selected. Besides, various decaying strategies of the pruning rate are adopted to stabilize the pruning process and improve the final performance as well. By gradually generating more identical filters within each cluster, ASCP can remove them through channel addition operation with almost no accuracy drop. Extensive experiments on CIFAR-10 and ImageNet datasets show that our method can achieve competitive results compared with many state-of-the-art algorithms.

25.Nucleus Segmentation and Analysis in Breast Cancer with the MIScnn Framework ⬇️

The NuCLS dataset contains over 220.000 annotations of cell nuclei in breast cancers. We show how to use these data to create a multi-rater model with the MIScnn Framework to automate the analysis of cell nuclei. For the model creation, we use the widespread U-Net approach embedded in a pipeline. This pipeline provides besides the high performance convolution neural network, several preprocessor techniques and a extended data exploration. The final model is tested in the evaluation phase using a wide variety of metrics with a subsequent visualization. Finally, the results are compared and interpreted with the results of the NuCLS study. As an outlook, indications are given which are important for the future development of models in the context of cell nuclei.

26.Level 2 Autonomous Driving on a Single Device: Diving into the Devils of Openpilot ⬇️

Equipped with a wide span of sensors, predominant autonomous driving solutions are becoming more modular-oriented for safe system design. Though these sensors have laid a solid foundation, most massive-production solutions up to date still fall into L2 phase. Among these, this http URL comes to our sight, claiming one $999 aftermarket device mounted with a single camera and board inside owns the ability to handle L2 scenarios. Together with open-sourced software of the entire system released by this http URL, the project is named Openpilot. Is it possible? If so, how is it made possible? With curiosity in mind, we deep-dive into Openpilot and conclude that its key to success is the end-to-end system design instead of a conventional modular framework. The model is briefed as Supercombo, and it can predict the ego vehicle's future trajectory and other road semantics on the fly from monocular input. Unfortunately, the training process and massive amount of data to make all these work are not publicly available. To achieve an intensive investigation, we try to reimplement the training details and test the pipeline on public benchmarks. The refactored network proposed in this work is referred to as OP-Deepdive. For a fair comparison of our version to the original Supercombo, we introduce a dual-model deployment scheme to test the driving performance in the real world. Experimental results on nuScenes, Comma2k19, CARLA, and in-house realistic scenarios verify that a low-cost device can indeed achieve most L2 functionalities and be on par with the original Supercombo model. In this report, we would like to share our latest findings, shed some light on the new perspective of end-to-end autonomous driving from an industrial product-level side, and potentially inspire the community to continue improving the performance. Our code, benchmarks are at this https URL.

27.RefCrowd: Grounding the Target in Crowd with Referring Expressions ⬇️

Crowd understanding has aroused the widespread interest in vision domain due to its important practical significance. Unfortunately, there is no effort to explore crowd understanding in multi-modal domain that bridges natural language and computer vision. Referring expression comprehension (REF) is such a representative multi-modal task. Current REF studies focus more on grounding the target object from multiple distinctive categories in general scenarios. It is difficult to applied to complex real-world crowd understanding. To fill this gap, we propose a new challenging dataset, called RefCrowd, which towards looking for the target person in crowd with referring expressions. It not only requires to sufficiently mine the natural language information, but also requires to carefully focus on subtle differences between the target and a crowd of persons with similar appearance, so as to realize the fine-grained mapping from language to vision. Furthermore, we propose a Fine-grained Multi-modal Attribute Contrastive Network (FMAC) to deal with REF in crowd understanding. It first decomposes the intricate visual and language features into attribute-aware multi-modal features, and then captures discriminative but robustness fine-grained attribute features to effectively distinguish these subtle differences between similar persons. The proposed method outperforms existing state-of-the-art (SoTA) methods on our RefCrowd dataset and existing REF datasets. In addition, we implement an end-to-end REF toolbox for the deeper research in multi-modal domain. Our dataset and code can be available at: \url{this https URL}.

28.K-Radar: 4D Radar Object Detection Dataset and Benchmark for Autonomous Driving in Various Weather Conditions ⬇️

Unlike RGB cameras that use visible light bands (384$\sim$769 THz) and Lidar that use infrared bands (361$\sim$331 THz), Radars use relatively longer wavelength radio bands (77$\sim$81 GHz), resulting in robust measurements in adverse weathers. Unfortunately, existing Radar datasets only contain a relatively small number of samples compared to the existing camera and Lidar datasets. This may hinder the development of sophisticated data-driven deep learning techniques for Radar-based perception. Moreover, most of the existing Radar datasets only provide 3D Radar tensor (3DRT) data that contain power measurements along the Doppler, range, and azimuth dimensions. As there is no elevation information, it is challenging to estimate the 3D bounding box of an object from 3DRT. In this work, we introduce KAIST-Radar (K-Radar), a novel large-scale object detection dataset and benchmark that contains 35K frames of 4D Radar tensor (4DRT) data with power measurements along the Doppler, range, azimuth, and elevation dimensions, together with carefully annotated 3D bounding box labels of objects on the roads. K-Radar includes challenging driving conditions such as adverse weathers (fog, rain, and snow) on various road structures (urban, suburban roads, alleyways, and highways). In addition to the 4DRT, we provide auxiliary measurements from carefully calibrated high-resolution Lidars, surround stereo cameras, and RTK-GPS. We also provide 4DRT-based object detection baseline neural networks (baseline NNs) and show that the height information is crucial for 3D object detection. And by comparing the baseline NN with a similarly-structured Lidar-based neural network, we demonstrate that 4D Radar is a more robust sensor for adverse weather conditions. All codes are available at this https URL.

29.Volumetric Supervised Contrastive Learning for Seismic Semantic Segmentation ⬇️

In seismic interpretation, pixel-level labels of various rock structures can be time-consuming and expensive to obtain. As a result, there oftentimes exists a non-trivial quantity of unlabeled data that is left unused simply because traditional deep learning methods rely on access to fully labeled volumes. To rectify this problem, contrastive learning approaches have been proposed that use a self-supervised methodology in order to learn useful representations from unlabeled data. However, traditional contrastive learning approaches are based on assumptions from the domain of natural images that do not make use of seismic context. In order to incorporate this context within contrastive learning, we propose a novel positive pair selection strategy based on the position of slices within a seismic volume. We show that the learnt representations from our method out-perform a state of the art contrastive learning methodology in a semantic segmentation task.

30.Zero-Shot Video Question Answering via Frozen Bidirectional Language Models ⬇️

Video question answering (VideoQA) is a complex task that requires diverse multi-modal data for training. Manual annotation of question and answers for videos, however, is tedious and prohibits scalability. To tackle this problem, recent methods consider zero-shot settings with no manual annotation of visual question-answer. In particular, a promising approach adapts frozen autoregressive language models pretrained on Web-scale text-only data to multi-modal inputs. In contrast, we here build on frozen bidirectional language models (BiLM) and show that such an approach provides a stronger and cheaper alternative for zero-shot VideoQA. In particular, (i) we combine visual inputs with the frozen BiLM using light trainable modules, (ii) we train such modules using Web-scraped multi-modal data, and finally (iii) we perform zero-shot VideoQA inference through masked language modeling, where the masked text is the answer to a given question. Our proposed approach, FrozenBiLM, outperforms the state of the art in zero-shot VideoQA by a significant margin on a variety of datasets, including LSMDC-FiB, iVQA, MSRVTT-QA, MSVD-QA, ActivityNet-QA, TGIF-FrameQA, How2QA and TVQA. It also demonstrates competitive performance in the few-shot and fully-supervised setting. Our code and models will be made publicly available at this https URL.

31.Self-Adaptive Label Augmentation for Semi-supervised Few-shot Classification ⬇️

Few-shot classification aims to learn a model that can generalize well to new tasks when only a few labeled samples are available. To make use of unlabeled data that are more abundantly available in real applications, Ren et al. \shortcite{ren2018meta} propose a semi-supervised few-shot classification method that assigns an appropriate label to each unlabeled sample by a manually defined metric. However, the manually defined metric fails to capture the intrinsic property in data. In this paper, we propose a \textbf{S}elf-\textbf{A}daptive \textbf{L}abel \textbf{A}ugmentation approach, called \textbf{SALA}, for semi-supervised few-shot classification. A major novelty of SALA is the task-adaptive metric, which can learn the metric adaptively for different tasks in an end-to-end fashion. Another appealing feature of SALA is a progressive neighbor selection strategy, which selects unlabeled data with high confidence progressively through the training phase. Experiments demonstrate that SALA outperforms several state-of-the-art methods for semi-supervised few-shot classification on benchmark datasets.

32.Trajectory-guided Control Prediction for End-to-end Autonomous Driving: A Simple yet Strong Baseline ⬇️

Current end-to-end autonomous driving methods either run a controller based on a planned trajectory or perform control prediction directly, which have spanned two separately studied lines of research. Seeing their potential mutual benefits to each other, this paper takes the initiative to explore the combination of these two well-developed worlds. Specifically, our integrated approach has two branches for trajectory planning and direct control, respectively. The trajectory branch predicts the future trajectory, while the control branch involves a novel multi-step prediction scheme such that the relationship between current actions and future states can be reasoned. The two branches are connected so that the control branch receives corresponding guidance from the trajectory branch at each time step. The outputs from two branches are then fused to achieve complementary advantages. Our results are evaluated in the closed-loop urban driving setting with challenging scenarios using the CARLA simulator. Even with a monocular camera input, the proposed approach ranks $first$ on the official CARLA Leaderboard, outperforming other complex candidates with multiple sensors or fusion mechanisms by a large margin. The source code and data will be made publicly available at this https URL.

33.Channel Importance Matters in Few-Shot Image Classification ⬇️

Few-Shot Learning (FSL) requires vision models to quickly adapt to brand-new classification tasks with a shift in task distribution. Understanding the difficulties posed by this task distribution shift is central to FSL. In this paper, we show that a simple channel-wise feature transformation may be the key to unraveling this secret from a channel perspective. When facing novel few-shot tasks in the test-time datasets, this transformation can greatly improve the generalization ability of learned image representations, while being agnostic to the choice of training algorithms and datasets. Through an in-depth analysis of this transformation, we find that the difficulty of representation transfer in FSL stems from the severe channel bias problem of image representations: channels may have different importance in different tasks, while convolutional neural networks are likely to be insensitive, or respond incorrectly to such a shift. This points out a core problem of the generalization ability of modern vision systems and needs further attention in the future.

34.A Simple Baseline for Adversarial Domain Adaptation-based Unsupervised Flood Forecasting ⬇️

Flood disasters cause enormous social and economic losses. However, both traditional physical models and learning-based flood forecasting models require massive historical flood data to train the model parameters. When come to some new site that does not have sufficient historical data, the model performance will drop dramatically due to overfitting. This technical report presents a Flood Domain Adaptation Network (FloodDAN), a baseline of applying Unsupervised Domain Adaptation (UDA) to the flood forecasting problem. Specifically, training of FloodDAN includes two stages: in the first stage, we train a rainfall encoder and a prediction head to learn general transferable hydrological knowledge on large-scale source domain data; in the second stage, we transfer the knowledge in the pretrained encoder into the rainfall encoder of target domain through adversarial domain alignment. During inference, we utilize the target domain rainfall encoder trained in the second stage and the prediction head trained in the first stage to get flood forecasting predictions. Experimental results on Tunxi and Changhua flood dataset show that FloodDAN can perform flood forecasting effectively with zero target domain supervision. The performance of the FloodDAN is on par with supervised models that uses 450-500 hours of supervision.

35.An Improved Normed-Deformable Convolution for Crowd Counting ⬇️

In recent years, crowd counting has become an important issue in computer vision. In most methods, the density maps are generated by convolving with a Gaussian kernel from the ground-truth dot maps which are marked around the center of human heads. Due to the fixed geometric structures in CNNs and indistinct head-scale information, the head features are obtained incompletely. Deformable convolution is proposed to exploit the scale-adaptive capabilities for CNN features in the heads. By learning the coordinate offsets of the sampling points, it is tractable to improve the ability to adjust the receptive field. However, the heads are not uniformly covered by the sampling points in the deformable convolution, resulting in loss of head information. To handle the non-uniformed sampling, an improved Normed-Deformable Convolution (\textit{i.e.,}NDConv) implemented by Normed-Deformable loss (\textit{i.e.,}NDloss) is proposed in this paper. The offsets of the sampling points which are constrained by NDloss tend to be more even. Then, the features in the heads are obtained more completely, leading to better performance. Especially, the proposed NDConv is a light-weight module which shares similar computation burden with Deformable Convolution. In the extensive experiments, our method outperforms state-of-the-art methods on ShanghaiTech A, ShanghaiTech B, UCF_QNRF, and UCF_CC_50 dataset, achieving 61.4, 7.8, 91.2, and 167.2 MAE, respectively. The code is available at this https URL

36.CARLANE: A Lane Detection Benchmark for Unsupervised Domain Adaptation from Simulation to multiple Real-World Domains ⬇️

Unsupervised Domain Adaptation demonstrates great potential to mitigate domain shifts by transferring models from labeled source domains to unlabeled target domains. While Unsupervised Domain Adaptation has been applied to a wide variety of complex vision tasks, only few works focus on lane detection for autonomous driving. This can be attributed to the lack of publicly available datasets. To facilitate research in these directions, we propose CARLANE, a 3-way sim-to-real domain adaptation benchmark for 2D lane detection. CARLANE encompasses the single-target datasets MoLane and TuLane and the multi-target dataset MuLane. These datasets are built from three different domains, which cover diverse scenes and contain a total of 163K unique images, 118K of which are annotated. In addition we evaluate and report systematic baselines, including our own method, which builds upon Prototypical Cross-domain Self-supervised Learning. We find that false positive and false negative rates of the evaluated domain adaptation methods are high compared to those of fully supervised baselines. This affirms the need for benchmarks such as CARLANE to further strengthen research in Unsupervised Domain Adaptation for lane detection. CARLANE, all evaluated models and the corresponding implementations are publicly available at this https URL.

37.DeepFormableTag: End-to-end Generation and Recognition of Deformable Fiducial Markers ⬇️

Fiducial markers have been broadly used to identify objects or embed messages that can be detected by a camera. Primarily, existing detection methods assume that markers are printed on ideally planar surfaces. Markers often fail to be recognized due to various imaging artifacts of optical/perspective distortion and motion blur. To overcome these limitations, we propose a novel deformable fiducial marker system that consists of three main parts: First, a fiducial marker generator creates a set of free-form color patterns to encode significantly large-scale information in unique visual codes. Second, a differentiable image simulator creates a training dataset of photorealistic scene images with the deformed markers, being rendered during optimization in a differentiable manner. The rendered images include realistic shading with specular reflection, optical distortion, defocus and motion blur, color alteration, imaging noise, and shape deformation of markers. Lastly, a trained marker detector seeks the regions of interest and recognizes multiple marker patterns simultaneously via inverse deformation transformation. The deformable marker creator and detector networks are jointly optimized via the differentiable photorealistic renderer in an end-to-end manner, allowing us to robustly recognize a wide range of deformable markers with high accuracy. Our deformable marker system is capable of decoding 36-bit messages successfully at ~29 fps with severe shape deformation. Results validate that our system significantly outperforms the traditional and data-driven marker methods. Our learning-based marker system opens up new interesting applications of fiducial markers, including cost-effective motion capture of the human body, active 3D scanning using our fiducial markers' array as structured light patterns, and robust augmented reality rendering of virtual objects on dynamic surfaces.

38.Backbones-Review: Feature Extraction Networks for Deep Learning and Deep Reinforcement Learning Approaches ⬇️

To understand the real world using various types of data, Artificial Intelligence (AI) is the most used technique nowadays. While finding the pattern within the analyzed data represents the main task. This is performed by extracting representative features step, which is proceeded using the statistical algorithms or using some specific filters. However, the selection of useful features from large-scale data represented a crucial challenge. Now, with the development of convolution neural networks (CNNs), the feature extraction operation has become more automatic and easier. CNNs allow to work on large-scale size of data, as well as cover different scenarios for a specific task. For computer vision tasks, convolutional networks are used to extract features also for the other parts of a deep learning model. The selection of a suitable network for feature extraction or the other parts of a DL model is not random work. So, the implementation of such a model can be related to the target task as well as the computational complexity of it. Many networks have been proposed and become the famous networks used for any DL models in any AI task. These networks are exploited for feature extraction or at the beginning of any DL model which is named backbones. A backbone is a known network trained in many other tasks before and demonstrates its effectiveness. In this paper, an overview of the existing backbones, e.g. VGGs, ResNets, DenseNet, etc, is given with a detailed description. Also, a couple of computer vision tasks are discussed by providing a review of each task regarding the backbones used. In addition, a comparison in terms of performance is also provided, based on the backbone used for each task.

39.Balancing Discriminability and Transferability for Source-Free Domain Adaptation ⬇️

Conventional domain adaptation (DA) techniques aim to improve domain transferability by learning domain-invariant representations; while concurrently preserving the task-discriminability knowledge gathered from the labeled source data. However, the requirement of simultaneous access to labeled source and unlabeled target renders them unsuitable for the challenging source-free DA setting. The trivial solution of realizing an effective original to generic domain mapping improves transferability but degrades task discriminability. Upon analyzing the hurdles from both theoretical and empirical standpoints, we derive novel insights to show that a mixup between original and corresponding translated generic samples enhances the discriminability-transferability trade-off while duly respecting the privacy-oriented source-free setting. A simple but effective realization of the proposed insights on top of the existing source-free DA approaches yields state-of-the-art performance with faster convergence. Beyond single-source, we also outperform multi-source prior-arts across both classification and semantic segmentation benchmarks.

40.Joint Class-Affinity Loss Correction for Robust Medical Image Segmentation with Noisy Labels ⬇️

Noisy labels collected with limited annotation cost prevent medical image segmentation algorithms from learning precise semantic correlations. Previous segmentation arts of learning with noisy labels merely perform a pixel-wise manner to preserve semantics, such as pixel-wise label correction, but neglect the pair-wise manner. In fact, we observe that the pair-wise manner capturing affinity relations between pixels can greatly reduce the label noise rate. Motivated by this observation, we present a novel perspective for noisy mitigation by incorporating both pixel-wise and pair-wise manners, where supervisions are derived from noisy class and affinity labels, respectively. Unifying the pixel-wise and pair-wise manners, we propose a robust Joint Class-Affinity Segmentation (JCAS) framework to combat label noise issues in medical image segmentation. Considering the affinity in pair-wise manner incorporates contextual dependencies, a differentiated affinity reasoning (DAR) module is devised to rectify the pixel-wise segmentation prediction by reasoning about intra-class and inter-class affinity relations. To further enhance the noise resistance, a class-affinity loss correction (CALC) strategy is designed to correct supervision signals via the modeled noise label distributions in class and affinity labels. Meanwhile, CALC strategy interacts the pixel-wise and pair-wise manners through the theoretically derived consistency regularization. Extensive experiments under both synthetic and real-world noisy labels corroborate the efficacy of the proposed JCAS framework with a minimum gap towards the upper bound performance. The source code is available at \url{this https URL}.

41.Patch-level Representation Learning for Self-supervised Vision Transformers ⬇️

Recent self-supervised learning (SSL) methods have shown impressive results in learning visual representations from unlabeled images. This paper aims to improve their performance further by utilizing the architectural advantages of the underlying neural network, as the current state-of-the-art visual pretext tasks for SSL do not enjoy the benefit, i.e., they are architecture-agnostic. In particular, we focus on Vision Transformers (ViTs), which have gained much attention recently as a better architectural choice, often outperforming convolutional networks for various visual tasks. The unique characteristic of ViT is that it takes a sequence of disjoint patches from an image and processes patch-level representations internally. Inspired by this, we design a simple yet effective visual pretext task, coined SelfPatch, for learning better patch-level representations. To be specific, we enforce invariance against each patch and its neighbors, i.e., each patch treats similar neighboring patches as positive samples. Consequently, training ViTs with SelfPatch learns more semantically meaningful relations among patches (without using human-annotated labels), which can be beneficial, in particular, to downstream tasks of a dense prediction type. Despite its simplicity, we demonstrate that it can significantly improve the performance of existing SSL methods for various visual tasks, including object detection and semantic segmentation. Specifically, SelfPatch significantly improves the recent self-supervised ViT, DINO, by achieving +1.3 AP on COCO object detection, +1.2 AP on COCO instance segmentation, and +2.9 mIoU on ADE20K semantic segmentation.

42.Image Captioning based on Feature Refinement and Reflective Decoding ⬇️

Automatically generating a description of an image in natural language is called image captioning. It is an active research topic that lies at the intersection of two major fields in artificial intelligence, computer vision, and natural language processing. Image captioning is one of the significant challenges in image understanding since it requires not only recognizing salient objects in the image but also their attributes and the way they interact. The system must then generate a syntactically and semantically correct caption that describes the image content in natural language. With the significant progress in deep learning models and their ability to effectively encode large sets of images and generate correct sentences, several neural-based captioning approaches have been proposed recently, each trying to achieve better accuracy and caption quality. This paper introduces an encoder-decoder-based image captioning system in which the encoder extracts spatial and global features for each region in the image using the Faster R-CNN with ResNet-101 as a backbone. This stage is followed by a refining model, which uses an attention-on-attention mechanism to extract the visual features of the target image objects, then determine their interactions. The decoder consists of an attention-based recurrent module and a reflective attention module, which collaboratively apply attention to the visual and textual features to enhance the decoder's ability to model long-term sequential dependencies. Extensive experiments performed on two benchmark datasets, MSCOCO and Flickr30K, show the effectiveness the proposed approach and the high quality of the generated captions.

43.Multi-scale Cooperative Multimodal Transformers for Multimodal Sentiment Analysis in Videos ⬇️

Multimodal sentiment analysis in videos is a key task in many real-world applications, which usually requires integrating multimodal streams including visual, verbal and acoustic behaviors. To improve the robustness of multimodal fusion, some of the existing methods let different modalities communicate with each other and modal the crossmodal interaction via transformers. However, these methods only use the single-scale representations during the interaction but forget to exploit multi-scale representations that contain different levels of semantic information. As a result, the representations learned by transformers could be biased especially for unaligned multimodal data. In this paper, we propose a multi-scale cooperative multimodal transformer (MCMulT) architecture for multimodal sentiment analysis. On the whole, the "multi-scale" mechanism is capable of exploiting the different levels of semantic information of each modality which are used for fine-grained crossmodal interactions. Meanwhile, each modality learns its feature hierarchies via integrating the crossmodal interactions from multiple level features of its source modality. In this way, each pair of modalities progressively builds feature hierarchies respectively in a cooperative manner. The empirical results illustrate that our MCMulT model not only outperforms existing approaches on unaligned multimodal sequences but also has strong performance on aligned multimodal sequences.

44.DreamNet: A Deep Riemannian Network based on SPD Manifold Learning for Visual Classification ⬇️

Image set-based visual classification methods have achieved remarkable performance, via characterising the image set in terms of a non-singular covariance matrix on a symmetric positive definite (SPD) manifold. To adapt to complicated visual scenarios better, several Riemannian networks (RiemNets) for SPD matrix nonlinear processing have recently been studied. However, it is pertinent to ask, whether greater accuracy gains can be achieved by simply increasing the depth of RiemNets. The answer appears to be negative, as deeper RiemNets tend to lose generalization ability. To explore a possible solution to this issue, we propose a new architecture for SPD matrix learning. Specifically, to enrich the deep representations, we adopt SPDNet [1] as the backbone, with a stacked Riemannian autoencoder (SRAE) built on the tail. The associated reconstruction error term can make the embedding functions of both SRAE and of each RAE an approximate identity mapping, which helps to prevent the degradation of statistical information. We then insert several residual-like blocks with shortcut connections to augment the representational capacity of SRAE, and to simplify the training of a deeper network. The experimental evidence demonstrates that our DreamNet can achieve improved accuracy with increased depth of the network.

45.A Simple Baseline for BEV Perception Without LiDAR ⬇️

Building 3D perception systems for autonomous vehicles that do not rely on LiDAR is a critical research problem because of the high expense of LiDAR systems compared to cameras and other sensors. Current methods use multi-view RGB data collected from cameras around the vehicle and neurally "lift" features from the perspective images to the 2D ground plane, yielding a "bird's eye view" (BEV) feature representation of the 3D space around the vehicle. Recent research focuses on the way the features are lifted from images to the BEV plane. We instead propose a simple baseline model, where the "lifting" step simply averages features from all projected image locations, and find that it outperforms the current state-of-the-art in BEV vehicle segmentation. Our ablations show that batch size, data augmentation, and input resolution play a large part in performance. Additionally, we reconsider the utility of radar input, which has previously been either ignored or found non-helpful by recent works. With a simple RGB-radar fusion module, we obtain a sizable boost in performance, approaching the accuracy of a LiDAR-enabled system.

46.Analysis and Extensions of Adversarial Training for Video Classification ⬇️

Adversarial training (AT) is a simple yet effective defense against adversarial attacks to image classification systems, which is based on augmenting the training set with attacks that maximize the loss. However, the effectiveness of AT as a defense for video classification has not been thoroughly studied. Our first contribution is to show that generating optimal attacks for video requires carefully tuning the attack parameters, especially the step size. Notably, we show that the optimal step size varies linearly with the attack budget. Our second contribution is to show that using a smaller (sub-optimal) attack budget at training time leads to a more robust performance at test time. Based on these findings, we propose three defenses against attacks with variable attack budgets. The first one, Adaptive AT, is a technique where the attack budget is drawn from a distribution that is adapted as training iterations proceed. The second, Curriculum AT, is a technique where the attack budget is increased as training iterations proceed. The third, Generative AT, further couples AT with a denoising generative adversarial network to boost robust performance. Experiments on the UCF101 dataset demonstrate that the proposed methods improve adversarial robustness against multiple attack types.

47.Technical Report for Argoverse2 Challenge 2022 -- Motion Forecasting Task ⬇️

We propose a motion forecasting model called BANet, which means Boundary-Aware Network, and it is a variant of LaneGCN. We believe that it is not enough to use only the lane centerline as input to obtain the embedding features of the vector map nodes. The lane centerline can only provide the topology of the lanes, and other elements of the vector map also contain rich information. For example, the lane boundary can provide traffic rule constraint information such as whether it is possible to change lanes which is very important. Therefore, we achieved better performance by encoding more vector map elements in the motion forecasting model.We report our results on the 2022 Argoverse2 Motion Forecasting challenge and rank 2nd on the test leaderboard.

48.Lifelong Wandering: A realistic few-shot online continual learning setting ⬇️

Online few-shot learning describes a setting where models are trained and evaluated on a stream of data while learning emerging classes. While prior work in this setting has achieved very promising performance on instance classification when learning from data-streams composed of a single indoor environment, we propose to extend this setting to consider object classification on a series of several indoor environments, which is likely to occur in applications such as robotics. Importantly, our setting, which we refer to as online few-shot continual learning, injects the well-studied issue of catastrophic forgetting into the few-shot online learning paradigm. In this work, we benchmark several existing methods and adapted baselines within our setting, and show there exists a trade-off between catastrophic forgetting and online performance. Our findings motivate the need for future work in this setting, which can achieve better online performance without catastrophic forgetting.

49.Dual Contrastive Attributed Graph Clustering Network ⬇️

Attributed graph clustering is one of the most important tasks in graph analysis field, the goal of which is to group nodes with similar representations into the same cluster without manual guidance. Recent studies based on graph contrastive learning have achieved impressive results in processing graph-structured data. However, existing graph contrastive learning based methods 1) do not directly address the clustering task, since the representation learning and clustering process are separated; 2) depend too much on graph data augmentation, which greatly limits the capability of contrastive learning; 3) ignore the contrastive message for subspace clustering. To accommodate the aforementioned issues, we propose a generic framework called Dual Contrastive Attributed Graph Clustering Network (DCAGC). In DCAGC, by leveraging Neighborhood Contrast Module, the similarity of the neighbor nodes will be maximized and the quality of the node representation will be improved. Meanwhile, the Contrastive Self-Expression Module is built by minimizing the node representation before and after the reconstruction of the self-expression layer to obtain a discriminative self-expression matrix for spectral clustering. All the modules of DCAGC are trained and optimized in a unified framework, so the learned node representation contains clustering-oriented messages. Extensive experimental results on four attributed graph datasets show the superiority of DCAGC compared with 16 state-of-the-art clustering methods. The code of this paper is available at this https URL.

50.PeQuENet: Perceptual Quality Enhancement of Compressed Video with Adaptation- and Attention-based Network ⬇️

In this paper we propose a generative adversarial network (GAN) framework to enhance the perceptual quality of compressed videos. Our framework includes attention and adaptation to different quantization parameters (QPs) in a single model. The attention module exploits global receptive fields that can capture and align long-range correlations between consecutive frames, which can be beneficial for enhancing perceptual quality of videos. The frame to be enhanced is fed into the deep network together with its neighboring frames, and in the first stage features at different depths are extracted. Then extracted features are fed into attention blocks to explore global temporal correlations, followed by a series of upsampling and convolution layers. Finally, the resulting features are processed by the QP-conditional adaptation module which leverages the corresponding QP information. In this way, a single model can be used to enhance adaptively to various QPs without requiring multiple models specific for every QP value, while having similar performance. Experimental results demonstrate the superior performance of the proposed PeQuENet compared with the state-of-the-art compressed video quality enhancement algorithms.

51.Improved surface reconstruction using high-frequency details ⬇️

Neural rendering can be used to reconstruct implicit representations of shapes without 3D supervision. However, current neural surface reconstruction methods have difficulty learning high-frequency details of shapes, so that the reconstructed shapes are often oversmoothed. We propose a novel method to improve the quality of surface reconstruction in neural rendering. We follow recent work to model surfaces as signed distance fields. First, we offer a derivation to analyze the relationship between the signed distance function, the volume density, the transparency function, and the weighting function used in the volume rendering equation. Second, we observe that attempting to jointly encode high-frequency and low frequency components in a single signed distance function leads to unstable optimization. We propose to decompose the signed distance function in a base function and a displacement function together with a coarse-to-fine strategy to gradually increase the high-frequency details. Finally, we propose to use an adaptive strategy that enables the optimization to focus on improving certain regions near the surface where the signed distance fields have artifacts. Our qualitative and quantitative results show that our method can reconstruct high-frequency surface details and obtain better surface reconstruction quality than the current state of the art. Code will be released at this https URL.

52.Action Spotting using Dense Detection Anchors Revisited: Submission to the SoccerNet Challenge 2022 ⬇️

This technical report describes our submission to the Action Spotting SoccerNet Challenge 2022. The challenge is part of the CVPR 2022 ActivityNet Workshop. Our submission is based on a method that we proposed recently, which focuses on increasing temporal precision via a densely sampled set of detection anchors. Due to its emphasis on temporal precision, this approach is able to produce competitive results on the tight average-mAP metric, which uses small temporal evaluation tolerances. This recently proposed metric is the evaluation criterion used for the challenge. In order to further improve results, here we introduce small changes in the pre- and post-processing steps, and also combine different input feature types via late fusion. This report describes the resulting overall approach, focusing on the modifications introduced. We also describe the training procedures used, and present our results.

53.Disentangling visual and written concepts in CLIP ⬇️

The CLIP network measures the similarity between natural text and images; in this work, we investigate the entanglement of the representation of word images and natural images in its image encoder. First, we find that the image encoder has an ability to match word images with natural images of scenes described by those words. This is consistent with previous research that suggests that the meaning and the spelling of a word might be entangled deep within the network. On the other hand, we also find that CLIP has a strong ability to match nonsense words, suggesting that processing of letters is separated from processing of their meaning. To explicitly determine whether the spelling capability of CLIP is separable, we devise a procedure for identifying representation subspaces that selectively isolate or eliminate spelling capabilities. We benchmark our methods against a range of retrieval tasks, and we also test them by measuring the appearance of text in CLIP-guided generated images. We find that our methods are able to cleanly separate spelling capabilities of CLIP from the visual processing of natural images.

54.What makes domain generalization hard? ⬇️

While several methodologies have been proposed for the daunting task of domain generalization, understanding what makes this task challenging has received little attention. Here we present SemanticDG (Semantic Domain Generalization): a benchmark with 15 photo-realistic domains with the same geometry, scene layout and camera parameters as the popular 3D ScanNet dataset, but with controlled domain shifts in lighting, materials, and viewpoints. Using this benchmark, we investigate the impact of each of these semantic shifts on generalization independently. Visual recognition models easily generalize to novel lighting, but struggle with distribution shifts in materials and viewpoints. Inspired by human vision, we hypothesize that scene context can serve as a bridge to help models generalize across material and viewpoint domain shifts and propose a context-aware vision transformer along with a contrastive loss over material and viewpoint changes to address these domain shifts. Our approach (dubbed as CDCNet) outperforms existing domain generalization methods by over an 18% margin. As a critical benchmark, we also conduct psychophysics experiments and find that humans generalize equally well across lighting, materials and viewpoints. The benchmark and computational model introduced here help understand the challenges associated with generalization across domains and provide initial steps towards extrapolation to semantic distribution shifts. We include all data and source code in the supplement.

55.Discrete Contrastive Diffusion for Cross-Modal and Conditional Generation ⬇️

Diffusion probabilistic models (DPMs) have become a popular approach to conditional generation, due to their promising results and support for cross-modal synthesis. A key desideratum in conditional synthesis is to achieve high correspondence between the conditioning input and generated output. Most existing methods learn such relationships implicitly, by incorporating the prior into the variational lower bound. In this work, we take a different route -- we enhance input-output connections by maximizing their mutual information using contrastive learning. To this end, we introduce a Conditional Discrete Contrastive Diffusion (CDCD) loss and design two contrastive diffusion mechanisms to effectively incorporate it into the denoising process. We formulate CDCD by connecting it with the conventional variational objectives. We demonstrate the efficacy of our approach in evaluations with three diverse, multimodal conditional synthesis tasks: dance-to-music generation, text-to-image synthesis, and class-conditioned image synthesis. On each, we achieve state-of-the-art or higher synthesis quality and improve the input-output correspondence. Furthermore, the proposed approach improves the convergence of diffusion models, reducing the number of required diffusion steps by more than 35% on two benchmarks, significantly increasing the inference speed.

56.SAVi++: Towards End-to-End Object-Centric Learning from Real-World Videos ⬇️

The visual world can be parsimoniously characterized in terms of distinct entities with sparse interactions. Discovering this compositional structure in dynamic visual scenes has proven challenging for end-to-end computer vision approaches unless explicit instance-level supervision is provided. Slot-based models leveraging motion cues have recently shown great promise in learning to represent, segment, and track objects without direct supervision, but they still fail to scale to complex real-world multi-object videos. In an effort to bridge this gap, we take inspiration from human development and hypothesize that information about scene geometry in the form of depth signals can facilitate object-centric learning. We introduce SAVi++, an object-centric video model which is trained to predict depth signals from a slot-based video representation. By further leveraging best practices for model scaling, we are able to train SAVi++ to segment complex dynamic scenes recorded with moving cameras, containing both static and moving objects of diverse appearance on naturalistic backgrounds, without the need for segmentation supervision. Finally, we demonstrate that by using sparse depth signals obtained from LiDAR, SAVi++ is able to learn emergent object segmentation and tracking from videos in the real-world Waymo Open dataset.

57.Boosting the Adversarial Transferability of Surrogate Model with Dark Knowledge ⬇️

Deep neural networks (DNNs) for image classification are known to be vulnerable to adversarial examples. And, the adversarial examples have transferability, which means an adversarial example for a DNN model can fool another black-box model with a non-trivial probability. This gave birth of the transfer-based adversarial attack where the adversarial examples generated by a pretrained or known model (called surrogate model) are used to conduct black-box attack. There are some work on how to generate the adversarial examples from a given surrogate model to achieve better transferability. However, training a special surrogate model to generate adversarial examples with better transferability is relatively under-explored. In this paper, we propose a method of training a surrogate model with abundant dark knowledge to boost the adversarial transferability of the adversarial examples generated by the surrogate model. This trained surrogate model is named dark surrogate model (DSM), and the proposed method to train DSM consists of two key components: a teacher model extracting dark knowledge and providing soft labels, and the mixing augmentation skill which enhances the dark knowledge of training data. Extensive experiments have been conducted to show that the proposed method can substantially improve the adversarial transferability of surrogate model across different architectures of surrogate model and optimizers for generating adversarial examples. We also show that the proposed method can be applied to other scenarios of transfer-based attack that contain dark knowledge, like face verification.

58.SoundSpaces 2.0: A Simulation Platform for Visual-Acoustic Learning ⬇️

We introduce SoundSpaces 2.0, a platform for on-the-fly geometry-based audio rendering for 3D environments. Given a 3D mesh of a real-world environment, SoundSpaces can generate highly realistic acoustics for arbitrary sounds captured from arbitrary microphone locations. Together with existing 3D visual assets, it supports an array of audio-visual research tasks, such as audio-visual navigation, mapping, source localization and separation, and acoustic matching. Compared to existing resources, SoundSpaces 2.0 has the advantages of allowing continuous spatial sampling, generalization to novel environments, and configurable microphone and material properties. To our best knowledge, this is the first geometry-based acoustic simulation that offers high fidelity and realism while also being fast enough to use for embodied learning. We showcase the simulator's properties and benchmark its performance against real-world audio measurements. In addition, through two downstream tasks covering embodied navigation and far-field automatic speech recognition, highlighting sim2real performance for the latter. SoundSpaces 2.0 is publicly available to facilitate wider research for perceptual systems that can both see and hear.

59.Deepfake histological images for enhancing digital pathology ⬇️

An optical microscopic examination of thinly cut stained tissue on glass slides prepared from a FFPE tissue blocks is the gold standard for tissue diagnostics. In addition, the diagnostic abilities and expertise of any pathologist is dependent on their direct experience with common as well as rarer variant morphologies. Recently, deep learning approaches have been used to successfully show a high level of accuracy for such tasks. However, obtaining expert-level annotated images is an expensive and time-consuming task and artificially synthesized histological images can prove greatly beneficial. Here, we present an approach to not only generate histological images that reproduce the diagnostic morphologic features of common disease but also provide a user ability to generate new and rare morphologies. Our approach involves developing a generative adversarial network model that synthesizes pathology images constrained by class labels. We investigated the ability of this framework in synthesizing realistic prostate and colon tissue images and assessed the utility of these images in augmenting diagnostic ability of machine learning methods as well as their usability by a panel of experienced anatomic pathologists. Synthetic data generated by our framework performed similar to real data in training a deep learning model for diagnosis. Pathologists were not able to distinguish between real and synthetic images and showed a similar level of inter-observer agreement for prostate cancer grading. We extended the approach to significantly more complex images from colon biopsies and showed that the complex microenvironment in such tissues can also be reproduced. Finally, we present the ability for a user to generate deepfake histological images via a simple markup of sematic labels.

60.Video Capsule Endoscopy Classification using Focal Modulation Guided Convolutional Neural Network ⬇️

Video capsule endoscopy is a hot topic in computer vision and medicine. Deep learning can have a positive impact on the future of video capsule endoscopy technology. It can improve the anomaly detection rate, reduce physicians' time for screening, and aid in real-world clinical analysis. CADx classification system for video capsule endoscopy has shown a great promise for further improvement. For example, detection of cancerous polyp and bleeding can lead to swift medical response and improve the survival rate of the patients. To this end, an automated CADx system must have high throughput and decent accuracy. In this paper, we propose FocalConvNet, a focal modulation network integrated with lightweight convolutional layers for the classification of small bowel anatomical landmarks and luminal findings. FocalConvNet leverages focal modulation to attain global context and allows global-local spatial interactions throughout the forward pass. Moreover, the convolutional block with its intrinsic inductive/learning bias and capacity to extract hierarchical features allows our FocalConvNet to achieve favourable results with high throughput. We compare our FocalConvNet with other SOTA on Kvasir-Capsule, a large-scale VCE dataset with 44,228 frames with 13 classes of different anomalies. Our proposed method achieves the weighted F1-score, recall and MCC} of 0.6734, 0.6373 and 0.2974, respectively outperforming other SOTA methodologies. Furthermore, we report the highest throughput of 148.02 images/second rate to establish the potential of FocalConvNet in a real-time clinical environment. The code of the proposed FocalConvNet is available at this https URL.

61.Longitudinal detection of new MS lesions using Deep Learning ⬇️

The detection of new multiple sclerosis (MS) lesions is an important marker of the evolution of the disease. The applicability of learning-based methods could automate this task efficiently. However, the lack of annotated longitudinal data with new-appearing lesions is a limiting factor for the training of robust and generalizing models. In this work, we describe a deep-learning-based pipeline addressing the challenging task of detecting and segmenting new MS lesions. First, we propose to use transfer-learning from a model trained on a segmentation task using single time-points. Therefore, we exploit knowledge from an easier task and for which more annotated datasets are available. Second, we propose a data synthesis strategy to generate realistic longitudinal time-points with new lesions using single time-point scans. In this way, we pretrain our detection model on large synthetic annotated datasets. Finally, we use a data-augmentation technique designed to simulate data diversity in MRI. By doing that, we increase the size of the available small annotated longitudinal datasets. Our ablation study showed that each contribution lead to an enhancement of the segmentation accuracy. Using the proposed pipeline, we obtained the best score for the segmentation and the detection of new MS lesions in the MSSEG2 MICCAI challenge.

62.Gradient-Based Adversarial and Out-of-Distribution Detection ⬇️

We propose to utilize gradients for detecting adversarial and out-of-distribution samples. We introduce confounding labels -- labels that differ from normal labels seen during training -- in gradient generation to probe the effective expressivity of neural networks. Gradients depict the amount of change required for a model to properly represent given inputs, providing insight into the representational power of the model established by network architectural properties as well as training data. By introducing a label of different design, we remove the dependency on ground truth labels for gradient generation during inference. We show that our gradient-based approach allows for capturing the anomaly in inputs based on the effective expressivity of the models with no hyperparameter tuning or additional processing, and outperforms state-of-the-art methods for adversarial and out-of-distribution detection.

63.Catastrophic overfitting is a bug but also a feature ⬇️

Despite clear computational advantages in building robust neural networks, adversarial training (AT) using single-step methods is unstable as it suffers from catastrophic overfitting (CO): Networks gain non-trivial robustness during the first stages of adversarial training, but suddenly reach a breaking point where they quickly lose all robustness in just a few iterations. Although some works have succeeded at preventing CO, the different mechanisms that lead to this remarkable failure mode are still poorly understood. In this work, however, we find that the interplay between the structure of the data and the dynamics of AT plays a fundamental role in CO. Specifically, through active interventions on typical datasets of natural images, we establish a causal link between the structure of the data and the onset of CO in single-step AT methods. This new perspective provides important insights into the mechanisms that lead to CO and paves the way towards a better understanding of the general dynamics of robust model construction. The code to reproduce the experiments of this paper can be found at this https URL .

64.A Closer Look at Smoothness in Domain Adversarial Training ⬇️

Domain adversarial training has been ubiquitous for achieving invariant representations and is used widely for various domain adaptation tasks. In recent times, methods converging to smooth optima have shown improved generalization for supervised learning tasks like classification. In this work, we analyze the effect of smoothness enhancing formulations on domain adversarial training, the objective of which is a combination of task loss (eg. classification, regression, etc.) and adversarial terms. We find that converging to a smooth minima with respect to (w.r.t.) task loss stabilizes the adversarial training leading to better performance on target domain. In contrast to task loss, our analysis shows that converging to smooth minima w.r.t. adversarial loss leads to sub-optimal generalization on the target domain. Based on the analysis, we introduce the Smooth Domain Adversarial Training (SDAT) procedure, which effectively enhances the performance of existing domain adversarial methods for both classification and object detection tasks. Our analysis also provides insight into the extensive usage of SGD over Adam in the community for domain adversarial training.

65.Lessons learned from the NeurIPS 2021 MetaDL challenge: Backbone fine-tuning without episodic meta-learning dominates for few-shot learning image classification ⬇️

Although deep neural networks are capable of achieving performance superior to humans on various tasks, they are notorious for requiring large amounts of data and computing resources, restricting their success to domains where such resources are available. Metalearning methods can address this problem by transferring knowledge from related tasks, thus reducing the amount of data and computing resources needed to learn new tasks. We organize the MetaDL competition series, which provide opportunities for research groups all over the world to create and experimentally assess new meta-(deep)learning solutions for real problems. In this paper, authored collaboratively between the competition organizers and the top-ranked participants, we describe the design of the competition, the datasets, the best experimental results, as well as the top-ranked methods in the NeurIPS 2021 challenge, which attracted 15 active teams who made it to the final phase (by outperforming the baseline), making over 100 code submissions during the feedback phase. The solutions of the top participants have been open-sourced. The lessons learned include that learning good representations is essential for effective transfer learning.

66.U-PET: MRI-based Dementia Detection with Joint Generation of Synthetic FDG-PET Images ⬇️

Alzheimer's disease (AD) is the most common cause of dementia. An early detection is crucial for slowing down the disease and mitigating risks related to the progression. While the combination of MRI and FDG-PET is the best image-based tool for diagnosis, FDG-PET is not always available. The reliable detection of Alzheimer's disease with only MRI could be beneficial, especially in regions where FDG-PET might not be affordable for all patients. To this end, we propose a multi-task method based on U-Net that takes T1-weighted MR images as an input to generate synthetic FDG-PET images and classifies the dementia progression of the patient into cognitive normal (CN), cognitive impairment (MCI), and AD. The attention gates used in both task heads can visualize the most relevant parts of the brain, guiding the examiner and adding interpretability. Results show the successful generation of synthetic FDG-PET images and a performance increase in disease classification over the naive single-task baseline.

67.Neural Scene Representation for Locomotion on Structured Terrain ⬇️

We propose a learning-based method to reconstruct the local terrain for locomotion with a mobile robot traversing urban environments. Using a stream of depth measurements from the onboard cameras and the robot's trajectory, the algorithm estimates the topography in the robot's vicinity. The raw measurements from these cameras are noisy and only provide partial and occluded observations that in many cases do not show the terrain the robot stands on. Therefore, we propose a 3D reconstruction model that faithfully reconstructs the scene, despite the noisy measurements and large amounts of missing data coming from the blind spots of the camera arrangement. The model consists of a 4D fully convolutional network on point clouds that learns the geometric priors to complete the scene from the context and an auto-regressive feedback to leverage spatio-temporal consistency and use evidence from the past. The network can be solely trained with synthetic data, and due to extensive augmentation, it is robust in the real world, as shown in the validation on a quadrupedal robot, ANYmal, traversing challenging settings. We run the pipeline on the robot's onboard low-power computer using an efficient sparse tensor implementation and show that the proposed method outperforms classical map representations.

68.Learning Effect of Lay People in Gesture-Based Locomotion in Virtual Reality ⬇️

Locomotion in Virtual Reality (VR) is an important part of VR applications. Many scientists are enriching the community with different variations that enable locomotion in VR. Some of the most promising methods are gesture-based and do not require additional handheld hardware. Recent work focused mostly on user preference and performance of the different locomotion techniques. This ignores the learning effect that users go through while new methods are being explored. In this work, it is investigated whether and how quickly users can adapt to a hand gesture-based locomotion system in VR. Four different locomotion techniques are implemented and tested by participants. The goal of this paper is twofold: First, it aims to encourage researchers to consider the learning effect in their studies. Second, this study aims to provide insight into the learning effect of users in gesture-based systems.

69.AMOS: A Large-Scale Abdominal Multi-Organ Benchmark for Versatile Medical Image Segmentation ⬇️

Despite the considerable progress in automatic abdominal multi-organ segmentation from CT/MRI scans in recent years, a comprehensive evaluation of the models' capabilities is hampered by the lack of a large-scale benchmark from diverse clinical scenarios. Constraint by the high cost of collecting and labeling 3D medical data, most of the deep learning models to date are driven by datasets with a limited number of organs of interest or samples, which still limits the power of modern deep models and makes it difficult to provide a fully comprehensive and fair estimate of various methods. To mitigate the limitations, we present AMOS, a large-scale, diverse, clinical dataset for abdominal organ segmentation. AMOS provides 500 CT and 100 MRI scans collected from multi-center, multi-vendor, multi-modality, multi-phase, multi-disease patients, each with voxel-level annotations of 15 abdominal organs, providing challenging examples and test-bed for studying robust segmentation algorithms under diverse targets and scenarios. We further benchmark several state-of-the-art medical segmentation models to evaluate the status of the existing methods on this new challenging dataset. We have made our datasets, benchmark servers, and baselines publicly available, and hope to inspire future research. Information can be found at this https URL.

70.Multi-View Imputation and Cross-Attention Network Based on Incomplete Longitudinal and Multi-Modal Data for Alzheimer's Disease Prediction ⬇️

Longitudinal variations and complementary information inherent in longitudinal and multi-modal data play an important role in Alzheimer's disease (AD) prediction, particularly in identifying subjects with mild cognitive impairment who are about to have AD. However, longitudinal and multi-modal data may have missing data, which hinders the effective application of these data. Additionally, previous longitudinal studies require existing longitudinal data to achieve prediction, but AD prediction is expected to be conducted at patients' baseline visit (BL) in clinical practice. Thus, we proposed a multi-view imputation and cross-attention network (MCNet) to integrate data imputation and AD prediction in a unified framework and achieve accurate AD prediction. First, a multi-view imputation method combined with adversarial learning, which can handle a wide range of missing data situations and reduce imputation errors, was presented. Second, two cross-attention blocks were introduced to exploit the potential associations in longitudinal and multi-modal data. Finally, a multi-task learning model was built for data imputation, longitudinal classification, and AD prediction tasks. When the model was properly trained, the disease progression information learned from longitudinal data can be leveraged by BL data to improve AD prediction. The proposed method was tested on two independent testing sets and single-model data at BL to verify its effectiveness and flexibility on AD prediction. Results showed that MCNet outperformed several state-of-the-art methods. Moreover, the interpretability of MCNet was presented. Thus, our MCNet is a tool with a great application potential in longitudinal and multi-modal data analysis for AD prediction. Codes are available at this https URL.

71.MoDi: Unconditional Motion Synthesis from Diverse Data ⬇️

The emergence of neural networks has revolutionized the field of motion synthesis. Yet, learning to unconditionally synthesize motions from a given distribution remains a challenging task, especially when the motions are highly diverse. We present MoDi, an unconditional generative model that synthesizes diverse motions. Our model is trained in a completely unsupervised setting from a diverse, unstructured and unlabeled motion dataset and yields a well-behaved, highly semantic latent space. The design of our model follows the prolific architecture of StyleGAN and adapts two of its key technical components into the motion domain: a set of style-codes injected into each level of the generator hierarchy and a mapping function that learns and forms a disentangled latent space. We show that despite the lack of any structure in the dataset, the latent space can be semantically clustered, and facilitates semantic editing and motion interpolation. In addition, we propose a technique to invert unseen motions into the latent space, and demonstrate latent-based motion editing operations that otherwise cannot be achieved by naive manipulation of explicit motion representations. Our qualitative and quantitative experiments show that our framework achieves state-of-the-art synthesis quality that can follow the distribution of highly diverse motion datasets. Code and trained models will be released at this https URL.

72.Multimodal Dialogue State Tracking ⬇️

Designed for tracking user goals in dialogues, a dialogue state tracker is an essential component in a dialogue system. However, the research of dialogue state tracking has largely been limited to unimodality, in which slots and slot values are limited by knowledge domains (e.g. restaurant domain with slots of restaurant name and price range) and are defined by specific database schema. In this paper, we propose to extend the definition of dialogue state tracking to multimodality. Specifically, we introduce a novel dialogue state tracking task to track the information of visual objects that are mentioned in video-grounded dialogues. Each new dialogue utterance may introduce a new video segment, new visual objects, or new object attributes, and a state tracker is required to update these information slots accordingly. We created a new synthetic benchmark and designed a novel baseline, Video-Dialogue Transformer Network (VDTN), for this task. VDTN combines both object-level features and segment-level features and learns contextual dependencies between videos and dialogues to generate multimodal dialogue states. We optimized VDTN for a state generation task as well as a self-supervised video understanding task which recovers video segment or object representations. Finally, we trained VDTN to use the decoded states in a response prediction task. Together with comprehensive ablation and qualitative analysis, we discovered interesting insights towards building more capable multimodal dialogue systems.

73.On Calibrated Model Uncertainty in Deep Learning ⬇️

Estimated uncertainty by approximate posteriors in Bayesian neural networks are prone to miscalibration, which leads to overconfident predictions in critical tasks that have a clear asymmetric cost or significant losses. Here, we extend the approximate inference for the loss-calibrated Bayesian framework to dropweights based Bayesian neural networks by maximising expected utility over a model posterior to calibrate uncertainty in deep learning. Furthermore, we show that decisions informed by loss-calibrated uncertainty can improve diagnostic performance to a greater extent than straightforward alternatives. We propose Maximum Uncertainty Calibration Error (MUCE) as a metric to measure calibrated confidence, in addition to its prediction especially for high-risk applications, where the goal is to minimise the worst-case deviation between error and estimated uncertainty. In experiments, we show the correlation between error in prediction and estimated uncertainty by interpreting Wasserstein distance as the accuracy of prediction. We evaluated the effectiveness of our approach to detecting Covid-19 from X-Ray images. Experimental results show that our method reduces miscalibration considerably, without impacting the models accuracy and improves reliability of computer-based diagnostics.

74.Reconstructing Training Data from Trained Neural Networks ⬇️

Understanding to what extent neural networks memorize training data is an intriguing question with practical and theoretical implications. In this paper we show that in some cases a significant fraction of the training data can in fact be reconstructed from the parameters of a trained neural network classifier. We propose a novel reconstruction scheme that stems from recent theoretical results about the implicit bias in training neural networks with gradient-based methods. To the best of our knowledge, our results are the first to show that reconstructing a large portion of the actual training samples from a trained neural network classifier is generally possible. This has negative implications on privacy, as it can be used as an attack for revealing sensitive training data. We demonstrate our method for binary MLP classifiers on a few standard computer vision datasets.

75.Edge Inference with Fully Differentiable Quantized Mixed Precision Neural Networks ⬇️

The large computing and memory cost of deep neural networks (DNNs) often precludes their use in resource-constrained devices. Quantizing the parameters and operations to lower bit-precision offers substantial memory and energy savings for neural network inference, facilitating the use of DNNs on edge computing platforms. Recent efforts at quantizing DNNs have employed a range of techniques encompassing progressive quantization, step-size adaptation, and gradient scaling. This paper proposes a new quantization approach for mixed precision convolutional neural networks (CNNs) targeting edge-computing. Our method establishes a new pareto frontier in model accuracy and memory footprint demonstrating a range of quantized models, delivering best-in-class accuracy below 4.3 MB of weights (wgts.) and activations (acts.). Our main contributions are: (i) hardware-aware heterogeneous differentiable quantization with tensor-sliced learned precision, (ii) targeted gradient modification for wgts. and acts. to mitigate quantization errors, and (iii) a multi-phase learning schedule to address instability in learning arising from updates to the learned quantizer and model parameters. We demonstrate the effectiveness of our techniques on the ImageNet dataset across a range of models including EfficientNet-Lite0 (e.g., 4.14MB of wgts. and acts. at 67.66% accuracy) and MobileNetV2 (e.g., 3.51MB wgts. and acts. at 65.39% accuracy).

76.Improving Diversity with Adversarially Learned Transformations for Domain Generalization ⬇️

To be successful in single source domain generalization, maximizing diversity of synthesized domains has emerged as one of the most effective strategies. Many of the recent successes have come from methods that pre-specify the types of diversity that a model is exposed to during training, so that it can ultimately generalize well to new domains. However, naïve diversity based augmentations do not work effectively for domain generalization either because they cannot model large domain shift, or because the span of transforms that are pre-specified do not cover the types of shift commonly occurring in domain generalization. To address this issue, we present a novel framework that uses adversarially learned transformations (ALT) using a neural network to model plausible, yet hard image transformations that fool the classifier. This network is randomly initialized for each batch and trained for a fixed number of steps to maximize classification error. Further, we enforce consistency between the classifier's predictions on the clean and transformed images. With extensive empirical analysis, we find that this new form of adversarial transformations achieve both objectives of diversity and hardness simultaneously, outperforming all existing techniques on competitive benchmarks for single source domain generalization. We also show that ALT can naturally work with existing diversity modules to produce highly distinct, and large transformations of the source domain leading to state-of-the-art performance.