Awesome

ICCV2021 Vector-Decomposed Disentanglement

Vector-Decomposed Disentanglement for Domain-Invariant Object Detection

Datasets

Daytime-Sunny, Dusk-Rainy, and Night-Rainy

[Download link]

[Models]

Based on BDD-100k dataset, we construct three datasets with different weather conditions. Particularly, Daytime-Sunny contains 27,708 daytime-sunny images. Dusk-Rainy contains 3,501 images. Night-Rainy contains 2,494 images.

Training

CUDA_VISIBLE_DEVICES=4 python da_train_net_gl.py --max_epochs 20 --cuda --dataset dc --dataset_t nr --bs 1 --da_use_contex --lc --gc

Testing

CUDA_VISIBLE_DEVICES=4 python test_SW_ICR_CCR_gl.py --cuda --modelname output_path --dataset nr --gc --lc --model_dir model_name

Citation

If you find this repository useful for your work, please cite as follows:

@article{wu2021vector,
  title={Vector-Decomposed Disentanglement for Domain-Invariant Object Detection},
  author={Wu, Aming and Liu, Rui and Han, Yahong and Zhu, Linchao and Yang, Yi},
  journal={IEEE International Conference on Computer Vision (ICCV)},
  year={2021}
}