Awesome Source-free Test-time Adaptation  

This is a curated list of research papers in Test-time Adaptation (TTA), which also goes by other names, such as Test-time Training (TTT), Source-free Domain Adaptation (SFDA) and Unsupervised Model Adaptation (UMA).

The repository is actively maintained. Pull requests or direct messages are welcome.

Table of Contents

• Methods
  • Self-supervision
  • Information Entropy
  • Batch Normalization
  • Pseudo Labeling
  • Class Prototype
  • Feature Alignment
  • Generative Modeling
  • Nearest Neighbors
  • Augmentation Invariance
  • Meta-learning
  • Time-varying
  • Others
• Benchmark
• Applications

Methods

Self-supervision

• Test-Time Training with Self-Supervision for Generalization under Distribution Shifts ICML'20 [Project]
• TTT++: When Does Self-Supervised Test-Time Training Fail or Thrive? NeurIPS'21 [Code]
• Model Adaptation: Historical Contrastive Learning for Unsupervised Domain Adaptation without Source Data NeurIPS'21
• Contrastive Test-Time Adaptation CVPR'22 [Code]
• Divide and Contrast: Source-free Domain Adaptation via Adaptive Contrastive Learning NeurIPS'22
• Test-Time Training with Masked Autoencoders NeurIPS'22 [Project]
• Improved Test-Time Adaptation for Domain Generalization CVPR'23 [Code]
• PromptStyler: Prompt-driven Style Generation for Source-free Domain Generalization ICCV'23
• MATE: Masked Autoencoders are Online 3D Test-Time Learners ICCV'23
• On the Robustness of Open-World Test-Time Training: Self-Training with Dynamic Prototype Expansion ICCV'23 [Code]

Information Entropy

• Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation ICML'20 [Code]
• Tent: Fully Test-Time Adaptation by Entropy Minimization ICLR'21 [Code]
• Uncertainty Reduction for Model Adaptation in Semantic Segmentation CVPR'21 [Code]
• Bayesian Adaptation for Covariate Shift NeurIPS'21
• Efficient Test-Time Model Adaptation without Forgetting ICML'22 [Code]
• Confidence Score for Source-Free Unsupervised Domain Adaptation ICML'22 [Code]
• Towards Stable Test-time Adaptation in Dynamic Wild World ICLR'23 [Code]

Batch Normalization

• Improving robustness against common corruptions by covariate shift adaptation NeurIPS'20 [Code]
• Tent: Fully Test-Time Adaptation by Entropy Minimization ICLR'21 [Code]
• Limitations of Post-Hoc Feature Alignment for Robustness CVPR'21 [Code]
• TTN: A Domain-Shift Aware Batch Normalization in Test-Time Adaptation ICLR'23
• Delta: Degradation-Free Fully Test-Time Adaptation ICLR'23
• Towards Stable Test-time Adaptation in Dynamic Wild World ICLR'23 [Code]
• Test-time Adaptation in the Dynamic World with Compound Domain Knowledge Management RA-L'23

Pseudo Labeling

• Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation ICML'20 [Code]
• Generative Pseudo-label Refinement for Unsupervised Domain Adaptation WACV'20
• A Free Lunch for Unsupervised Domain Adaptive Object Detection without Source Data AAAI'21
• Uncertainty Reduction for Model Adaptation in Semantic Segmentation CVPR'21 [Code]
• Adapting ImageNet-scale models to complex distribution shifts with self-learning TMLR'22 [Code]
• Continual Test-Time Domain Adaptation CVPR'22 [Code]
• Contrastive Test-Time Adaptation CVPR'22 [Code]
• Test-Time Adaptation via Conjugate Pseudo-labels NeurIPS'22 [Code]
• Towards Understanding GD with Hard and Conjugate Pseudo-labels for Test-Time Adaptation ICLR'23
• TeSLA: Test-Time Self-Learning With Automatic Adversarial Augmentation CVPR'23 [Code]

Class Prototype

• Model Adaptation: Unsupervised Domain Adaptation Without Source Data CVPR'20
• Test-Time Classifier Adjustment Module for Model-Agnostic Domain Generalization NeurIPS'21 [Code]
• Revisiting Realistic Test-Time Training: Sequential Inference and Adaptation by Anchored Clustering NeurIPS'22 [Code]
• Attracting and Dispersing: A Simple Approach for Source-free Domain Adaptation NeurIPS'22 [Code]
• On the Robustness of Open-World Test-Time Training: Self-Training with Dynamic Prototype Expansion ICCV'23 [Code]

Feature Alignment

• SoFA: Source-data-free Feature Alignment for Unsupervised Domain Adaptation WACV'21
• Adaptive Adversarial Network for Source-Free Domain Adaptation ICCV'21
• TTT++: When Does Self-Supervised Test-Time Training Fail or Thrive? NeurIPS'21 [Code]
• Source-Free Adaptation to Measurement Shift via Bottom-Up Feature Restoration ICLR'22 [Code]
• Invariance Through Latent Alignment RSS'22 [Project]
• Source-Free Domain Adaptation via Distribution Estimation CVPR'22
• ActMAD: Activation Matching to Align Distributions for Test-Time-Training CVPR'23 [Code]
• Feature Alignment and Uniformity for Test Time Adaptation CVPR'23 [Code]
• Robustness to corruption in pre-trained Bayesian neural networks ICLR'23

Generative Modeling

• Model Adaptation: Unsupervised Domain Adaptation without Source Data CVPR'20
• Domain Impression: A Source Data Free Domain Adaptation Method WACV'21
• Back to the Source: Diffusion-Driven Test-Time Adaptation CVPR'23 [Code]

Nearest Neighbors

• Generalized Source-free Domain Adaptation ICCV'21 [Code]
• Exploiting the Intrinsic Neighborhood Structure for Source-free Domain Adaptation NeurIPS'21 [Code]
• Test-Time Adaptation via Self-Training with Nearest Neighbor Information ICLR'23 [Code]

Augmentation Invariance

• MEMO: Test Time Robustness via Adaptation and Augmentation NeurIPS'22 [Code]
• Test time Adaptation through Perturbation Robustness NeurIPS-WS'21
• Balancing Discriminability and Transferability for Source-Free Domain Adaptation ICML'22 [Project]

Meta-learning

• Test-Time Fast Adaptation for Dynamic Scene Deblurring via Meta-Auxiliary Learning CVPR'21
• Adaptive Risk Minimization: Learning to Adapt to Domain Shift NeurIPS'21 [Code]
• Learning to Generalize across Domains on Single Test Samples ICLR'22 [Code]
• Meta-DMoE: Adapting to Domain Shift by Meta-Distillation from Mixture-of-Experts NeurIPS'22 [Code]

Time-varying

• Continual Test-Time Domain Adaptation CVPR'22 [Code]
• NOTE: Robust Continual Test-time Adaptation Against Temporal Correlation NeurIPS'22 [Code]
• Extrapolative Continuous-time Bayesian Neural Network for Fast Training-free Test-time Adaptation NeurIPS'22 [Code]
• Decorate the Newcomers: Visual Domain Prompt for Continual Test Time Adaptation AAAI'23
• Robust Test-Time Adaptation in Dynamic Scenarios CVPR'23 [Code]
• A Probabilistic Framework for Lifelong Test-Time Adaptation CVPR'23 [Code]
• Robust Mean Teacher for Continual and Gradual Test-Time Adaptation CVPR'23 [Code]
• EcoTTA: Memory-Efficient Continual Test-time Adaptation via Self-distilled Regularization CVPR'23 [Project]

Others

• Domain Adaptation in the Absence of Source Domain Data KDD'16
• Semantic Photo Manipulation with a Generative Image Prior SIGGRAPH'19
• Collaborative Sampling in Generative Adversarial Networks AAAI'20 [Code]
• Universal Source-Free Domain Adaptation CVPR'20 [Project]
• Adaptive Methods for Real-World Domain Generalization CVPR'21 [Code]
• Parameter-free Online Test-time Adaptation CVPR'22 [Code]
• Evaluating the Adversarial Robustness of Adaptive Test-time Defenses ICML'22
• MECTA: Memory-Economic Continual Test-Time Model Adaptation ICLR'23
• Neuro-Modulated Hebbian Learning for Fully Test-Time Adaptation CVPR'23

Benchmark

• On Pitfalls of Test-Time Adaptation ICML'2023 [Code]

Applications

• Consistent Video Depth Estimation SIGGRAPH'2020 [Project]
• Self-Supervised Policy Adaptation during Deployment ICLR'21 [Project]
• Source-Free Domain Adaptation for Image Segmentation MICCAI'20 [Code]
• Fully Test-Time Adaptation for Image Segmentation MICCAI'21
• Adapting Off-the-Shelf Source Segmenter for Target Medical Image Segmentation MICCAI'21
• Source-Free Domain Adaptation for Semantic Segmentation CVPR'21
• Generalize Then Adapt: Source-Free Domain Adaptive Semantic Segmentation ICCV'21 [Project]
• SS-SFDA: Self-Supervised Source-Free Domain Adaptation for Road Segmentation in Hazardous Environments ICCV'21 [Project]
• Test-Time Personalization with a Transformer for Human Pose Estimation NeurIPS'21 [Code]
• Towards Robust and Adaptive Motion Forecasting: A Causal Representation Perspective CVPR'22 [Code]
• Ev-TTA: Test-Time Adaptation for Event-Based Object Recognition CVPR'22 [Code]
• MM-TTA: Multi-Modal Test-Time Adaptation for 3D Semantic Segmentation CVPR'22 [Project]
• Source-Free Object Detection by Learning to Overlook Domain Style CVPR'22 [Code]
• On the Road to Online Adaptation for Semantic Image Segmentation CVPR'22
• Test-Time Training Can Close the Natural Distribution Shift Performance Gap in Deep Learning Based Compressed Sensing ICML'22 [Code]
• The Dual Form of Neural Networks Revisited: Connecting Test Time Predictions to Training Patterns via Spotlights of Attention ICML'22 [Code]
• Source-free Video Domain Adaptation by Learning Temporal Consistency for Action Recognition ECCV'22 [Code]
• AuxAdapt: Stable and Efficient Test-Time Adaptation for Temporally Consistent Video Semantic Segmentation WACV'22
• Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models NeurIPS'22 [Project]
• TTA-COPE: Test-Time Adaptation for Category-Level Object Pose Estimation CVPR'23 [Project]
• SfM-TTR: Using Structure from Motion for Test-Time Refinement of Single-View Depth Networks CVPR'23 [Code]
• Video Test-Time Adaptation for Action Recognition CVPR'23 [Project]