This is the official PyTorch implementation of our WAS paper:
Weakly Supervised Assistance for Unsupervised Domain Adaptive Object Detection
Xinhe Ouyang, Yanting Pei, Minhao Hao, Fan Yang
In this repository, we include the implementation of Contrastive Mean Teacher, integrated with both base methods Contrastive Mean Teacher (CMT, [code] [paper]) and H2FA (H2FA, [code] [paper]). Our code is based on the publicly available implementation of these two methods.
We follow CMT and H2FA original instructions to set up the environment and datasets. The details are included in the README files.
Here is an example script for reproducing our results of WAS on Cityscapes -> Foggy Cityscapes (all splits):
#prepare pseudo_labels(these pseudo-labels are generate by CMT,
#if you want to try other UDAOD pseudo-labels you can construct yours)
#I have provide foggy2city_train_pseudo_label.json and voc2clipart_train_pseudo_label.json in WAS/pseudo_labels
# enter the code directory for WAS
cd UDAOD/h2fa
# activate AT environment
conda activate h2fa
python train_net.py \
--config-file ../configs/CrossDomain-Detection/h2fa_rcnn_R_101_DC5_foggycityscapes.yaml \
--num-gpus 1 SOLVER.IMS_PER_BATCH 2 SOLVER.BASE_LR 0.0025 \
SOLVER.STEPS 48000,64000 SOLVER.MAX_ITER 72000
- Other configuration options may be found in
configs. - To resume the training, simply add
--resumeto the command. - To evaluate an existing model checkpoint, add
--eval-onlyand specifyMODEL.WEIGHTS path/to/your/weights.pthin the command.
Here we list the model weights for the results included in our paper:
| Dataset | Method | mAP (AP50) | Weights |
|---|---|---|---|
| Cityscapes -> Foggy Cityscapes | CMT + WAS | 55.1 | further |
| Pascal VOC -> Clipart1k | CMT + WAS | 48.1 | further |
This project is released under the Apache 2.0 license. Other codes from open source repository follows the original distributive licenses.