WSRD-DNSR

This is the repository of the WSRD dataset, used as benchmark for the NTIRE2023 Challenge on Image Shadow Removal.

Training data

The collection of 1000 roughly aligned image pairs can be found using the following URLS: train input link | train gt link.

Validation data

We provide a database of 100 images, that were used for evaluation on the provided Codalab server: valid input link | valid gt link.

Testing data

The input images used for testing in the Final Phase of the challenge are available here. Since we are planning for a second edition of the challenge, the test ground-truth images will remain private, for the moment. Follow the current repository for more updates.

DNSR

Along with the provided data, we also provide results for DNSR, a baseline optimized for reconstruction fidelity.

• Checkpoints for DNSR are available here.
• Results of the DNSR over the WSRD benchmark are available here.
• Additional results for the ISTD/ISTD+ benchmarks are available here.

Please follow the current repository for more updates.