AQT: Adversarial Query Transformers for Domain Adaptive Object Detection

By Wei-Jie Huang, Yu-Lin Lu, Shih-Yao Lin, Yusheng Xie, and Yen-Yu Lin.

This repository contains the implementation accompanying our paper AQT: Adversarial Query Transformers for Domain Adaptive Object Detection. This work was accepted to IJCAI-ECAI 2022.

If you find it helpful for your research, please consider citing:

@inproceedings{huang2022aqt,
  title     = {AQT: Adversarial Query Transformers for Domain Adaptive Object Detection},
  author    = {Huang, Wei-Jie and Lu, Yu-Lin and Lin, Shih-Yao and Xie, Yusheng and Lin, Yen-Yu},
  booktitle = {International Joint Conference on Artificial Intelligence (IJCAI)},
  year      = {2022},
}

Acknowledgment

This implementation is bulit upon Deformable DETR.

Installation

Please refer to the instructions here. We leave our system information for reference.

OS: Ubuntu 16.04
Python: 3.6
CUDA: 9.2
cuDNN: 7
PyTorch: 1.5.1
torchvision: 0.6.1

Dataset Preparation

Please construct the datasets following these steps:

Download the datasets from their sources:
- Cityscapes / Foggy Cityscapes: download gtFine_trainvaltest.zip (labels), leftImg8bit_trainvaltest.zip (Cityscapes images), leftImg8bit_trainvaltest_foggy.zip (Foggy Cityscapes images) from the official website.
- Sim10k: download images and annotations of 10k subset from the official website.
- BDD100k: download 100k Images and Labels from the official website.
Convert the annotation files into COCO-format annotations (you can build them following the annotation conversion script).
Prepare a directory datasets as follows (or you can modify the setting here)

datasets/
├─ bdd_daytime/
│  ├─ annotations/
│  ├─ train/
│  ├─ val/
├─ cityscapes/
│  ├─ annotations/
│  ├─ leftImg8bit/
│  |  ├─ train/
│  |  ├─ val/
│  ├─ leftImg8bit_foggy/
│  |  ├─ train/
│  |  ├─ val/
├─ sim10k/
│  ├─ annotations/
│  ├─ VOC2012/

Training / Evaluation

We provide training script on single node as follows, please refer to the instructions here for other settings.

GPUS_PER_NODE={NUM_GPUS} ./tools/run_dist_launch.sh {NUM_GPUS} python main.py --config_file {CONFIG_FILE}

We use yacs for configuration. The configuration files are in ./configs. If you want to override configuration using command line arguments, please consider the following script, which performs evaluation on a pre-trained model:

GPUS_PER_NODE={NUM_GPUS} ./tools/run_dist_launch.sh {NUM_GPUS} python main.py --config_file {CONFIG_FILE} --opts EVAL True RESUME {CHECKPOINT_FILE}

Pre-trained models

Cityscapes to Foggy Cityscapes: cfg, model
Sim10k to Cityscapes: cfg, model
Cityscapes to BDD100k daytime: cfg, model

weii41392 / AQT