A Pytorch Implementation of Unbiased Mean Teacher for Cross-domain Object Detection (CVPR 2021)

Introduction

Follow the implementation of faster-rcnn.pytorch to set up the environment. In our implementation, we use Pytorch 0.4.0 on a single GeForce GTX 1080 Ti.

Environment Preparation

Data Preparation

Please follow the instructions in DA_detection to prepare PASCAL_VOC 07+12, Clipart1k, WaterColor2k, and SIM10K. We use CycleGAN to generate the source/target-like images.

All the data arrangements follow the format of PASCAL_VOC. Our dataset config system also follow the DA_detection.

Train

 CUDA_VISIBLE_DEVICES=$GPU_ID python umt_train.py \
                    --dataset {SOURCE DATASET} --dataset_t {Target DATASET} --net {vgg16 or res101}

Taking clipart as an example:

 CUDA_VISIBLE_DEVICES=$GPU_ID python umt_train.py \
                    --dataset pascal_voc_07_12 --dataset_t clipart --net res101

Test

./test.sh {GUP_ID} {MODEL_PATH}

Citation

Please cite the following reference if you utilize this repository for your project.

@inproceedings{deng2021unbiased,
  title={Unbiased Mean Teacher for Cross-Domain Object Detection},
  author={Deng, Jinhong and Li, Wen and Chen, Yuhua and Duan, Lixin},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={4091--4101},
  year={2021}
}