CAT: Exploiting Inter-Class Dynamics for Domain Adaptive Object Detection

This repo is the official implementation of our paper: CAT: Exploiting Inter-Class Dynamics for Domain Adaptive Object Detection Mikhail Kennerley, Jian-Gang Wang, Bharadwaj Veeravalli, Robby T. Tan IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024 [Paper]

Installation

Prerequisites

• Python ≥ 3.6
• PyTorch ≥ 1.5 and torchvision that matches the PyTorch installation.
• Detectron2 == 0.6

Our tested environment

• 4 3090 GPUs

VGG-16

For the VGG backbone, we use the converted weights from CMT. Please put this file at checkpoints/vgg16_bn-6c64b313_converted.pth.

Dataset download

1. Download the Cityscapes dataset

2. Organize the dataset following:

adaptive_teacher/
└── datasets/
    └── cityscapes/
        ├── gtFine/
            ├── train/
            └── test/
            └── val/
        ├── leftImg8bit/
            ├── train/
            └── test/
            └── val/
   └── cityscapes_foggy/
        ├── gtFine/
            ├── train/
            └── test/
            └── val/
        ├── leftImg8bit/
            ├── train/
            └── test/
            └── val/

Training

• Train the Class-Aware Teacher under cityscapes (source) and foggy cityscapes (target)

python train_net.py\
      --num-gpus 4\
      --config configs/faster_rcnn_VGG_cross_city.yaml\
      OUTPUT_DIR output/exp_city

Citation

If you use Class Aware Teacher in your research or wish to refer to the results published in our paper, please use the following BibTeX entry:

@inproceedings{kennerley2024cat,
  title={CAT: Exploiting Inter-Class Dynamics for Domain Adaptive Object Detection},
  author={Mikhail Kennerley, Jian-Gang Wang, Bharadwaj Veeravalli, Robby T. Tan},
  booktitle={2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2024},
}

Acknowledgements

Code is adapted from Detectron2 and Adaptive Teacher.