BlenDA: Domain Adaptive Object Detection through diffusion-based blending

ICASSP 2024

Paper | arXiv

Tzuhsuan Huang*, Chen-Che Huang*, Chung-Hao Ku, Jun-Cheng Chen

Environment setup

git clone https://github.com/Jason-user/BlenDA

cd BlenDA/

conda env create -f environment.yml

conda activate blenda

pip install -r requirements.txt

cd models/ops/
python setup.py build install

Download pre-trained weight

Please refer to AQT
After downloading, put it into the folder called released_weights

BlenDA
├── released_weights
|   ├── cityscapes_to_bdd100k_daytime.pth
|   └── cityscapes_to_foggy_cityscapes.pth

Datasets

Please refer to Ciityscapes and BDD100K
To generate city_instruction_cityfoggy and city_instruction_bdd, you can refer to InstructPix2Pix

BlenDA
├── data/
|   ├── bdd_daytime/
|   |   ├── annotations/
|   |   ├── train/
|   |   └──  val/
|   ├── cityscapes/
|   |   ├── annotations/
|   |   └── leftImg8bit/
|   |       ├── train/
|   |       └── val/
|   ├── cityscapes_foggy
|   |   ├── annotations/
|   |   └── leftImg8bit/
|   |       ├── train/
|   |       └── val/
|   ├── city_instruction_cityfoggy/
|   |   ├── train/
|   |   └── val/
|   └── city_instruction_bdd/
|       ├── train/
|       └── val/

Training

cd exps/

Cityscapes to BDD100K

cd c2b/
bash run.sh

Cityscapes to Foggy Cityscapes

cd c2fc/
bash run.sh