FDRNet

Code for our ICCV 2021 paper "Mitigating Intensity Bias in Shadow Detection via Feature Decomposition and Reweighting"

How to Use

create conda environment

conda env create -f env.yaml

To use CRF refinement, you will need to mannually install pydensecrf.

WARNING: To reproduce the results reported in our paper, please make sure major pacakges (pytorch, opencv, etc.) are with the same version speficified in env.yaml.

run inference

• download the checkpoint from here.
• specify data_root, and run python test.py.
• run python crf_refine.py.
• check the results w/ and w/o CRF refinement in test/raw and test/crf respectively

Results

BER scores are specified below.

You can access qualitative reseults from BaiduNetDisk (passcode:4j3i).

TODOs

• [ ] move the logic of brightness shift to the dataset class; rewrite dataset class.
• [ ] remove feature extraction hook, use segmentation_models.pytorch encoder instead.
• [ ] use timm's register_model decorator