README.md

NTIRE2023 Image Shadow Removal Challenge Team IIM_TTI

This repository is a participation code for the Image Shadow Removal Challenge of the NTIRE 2023 challenges.

Datasets

datasets were distributed to those who participated in the competition. Competition page

Requirement

• Python>=3.7

pip install -r requirements.txt

Inference

1. Arrange the final test data as follows:

datasets
└── official
    └── test_final
        └── input
            ├── 0000.png
            ├── 0001.png
            ├── ...
            └── 0099.png

2. Download weights.

Download Link

Place the files in the 'weights' directory as follows:

weights/shadow_former+.pth

3. Run the inference.

python inference.py \
  --weights weights/shadow_former+.pth \
  --result_dir results/shadow_former+ \
  --joint_learning_alpha 1 \
  --gpus 0

The results will be output to results/shadow_former+.

Training preprocessing

For the training preprocesing, our proposed method uses SASMA.
The code for SASMA is SASMA.ipynb. For more information on SASMA, see the Citation section.

References

• ShadowFormer
• MTMT