Masked AutoDecoder

This is the official implementation of the paper "Masked AutoDecoder is Effective Multi-Task Vision Generalist"

Authors: Han Qiu, Jiaxing Huang, Peng Gao, Lewei Lu, Xiaoqin Zhang, Shijian Lu

In this work, we design Masked AutoDecoder(MAD), an effective multi-task vision generalist. MAD consists of two core designs. First, we develop a parallel decoding framework that introduces bi-directional attention to capture contextual dependencies comprehensively and decode vision task sequences in parallel. Second, we design a masked sequence modeling approach that learns rich task contexts by masking and reconstructing task sequences. In this way, MAD handles all the tasks by a single network branch and a simple cross-entropy loss with minimal task-specific designs.

Installation

First Install Detectron2.

Then,

cd MAD
pip install -e .

Please refer to Installation Instructions of Detrex for details of installation.

Data Preparation

# First prepare COCO dataset at "./datasets" as following:
- datasets
  - coco
    - annotation
      - captions_train2017.json
      - captions_val2017.json
      - instances_train2017.json
      - instances_val2017.json
      - person_keypoints_train2017.json
      - person_keypoints_val2017.json
    - train2017
    - val2017

# merge the keypoint anno and coco instance anno 
python ./project/mad/data/merge_annotations.py

Training

python tools/train_net.py --num-gpus 8 --dist-url auto --config-file ./project/mad/model/config.py

Evaluation

python tools/train_net.py --num-gpus 1 --dist-url auto --config-file ./project/mad/model/config.py --eval-only 

Acknowledgement

We build MAD based on detrex.

Citation

If you find our work helpful please cite:

@InProceedings{Qiu_2024_CVPR,
    author    = {Qiu, Han and Huang, Jiaxing and Gao, Peng and Lu, Lewei and Zhang, Xiaoqin and Lu, Shijian},
    title     = {Masked AutoDecoder is Effective Multi-Task Vision Generalist},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024},
    pages     = {14152-14161}
}