This is the official implementation of the TNNLS 2023 paper "Hue Guidance Network for Single Image Reflection Removal".
-
Prepare the training data as follows:
|--training_data |--JPEGImages # VOC2012 images for synthesis the reflection data |--real_train # real world reflection images and corresponding GT from previous methods, eg, real, nature |--blended # reflection images |--transmission_layer # ground truth (GT)
-
training phase
python train_HGRR.py --name train_DB_wboosting --hyper --inet HGRR_wboosting1128 --lambda_vgg 0.02 --lambda_newloss 50 --lambda_newloss_H 20
-
evaluation phase
python test_HGRR.py --name test -r --inet HGRR_wboosting1128 --icnn_path best.pt --hyper
Results of Reflection Benchmarks
If you have any problem, please feel free to contact me (zyr@mail.ustc.edu.cn).
- Our codes are inspired by ERRNet