The implementation of our CVPR2022 paper "Cross Domain Object Detection by Target-Perceived Dual Branch Distillation"
Paper: https://arxiv.org/pdf/2205.01291
2022.5.2
update the test code and model.
2022.5.10
update the train code and model.\
The enviroment and data preparation will be updated soon.
Inference:
python train_net.py --eval-only --num-gpus 8 --config-file configs/TDD/***.yaml MODEL.WEIGHTS ***.pth
Released Weights
For your convenience, the trained models are provided in this link. (BaiduYun code: e0lh)
Joint Pretrain:
The first stage of pretraining our detector with source images and target-like images.
python train_net.py --num-gpus 8 --config-file configs/prertain_r50_FPN/faster_rcnn_R_50_FPN_focal_cross_bdd.yaml
The weights are also preovided in the link above.
python train_net.py --num-gpus 8 --eval-only --config-file *****.yaml MODEL.WEIGHTS ********.pth\
TDD Train
python train_net.py --num-gpus 8 --config-file configs/TDD/faster_rcnn_R_50_FPN_foggy.yaml\
Note: Now only support that one gpu two images ( when GPUS=8, set IMG_PER_BATCH_LABEL: 8 IMG_PER_BATCH_UNLABEL: 8)
The two stage can be trained with one config file by modify the BURN_UP_STEP which means the step of joint pretrain.