search

Yang-Liu1082 / InvDN

Implementation for the paper: Invertible Denoising Network: A Light Solution for Real Noise Removal (CVPR2021).
Apache License 2.0
185 stars 35 forks source link

What's the pre-processing of DIV2K for training? missing contents in crop_***.py #3

Closed Zirconium2159 closed 3 years ago

Zirconium2159 commented 3 years ago

I tried to perform training on DIV2K dataset with the help of LQGTSN_dataset.py. However, the performance seems poor(under 30dB PSNR, quite lower than proposed). The loss curves also differ a lot from the training on SIDD dataset. Maybe I missed some important configurations when training on the DIV2K. It will be appreciated if you provide corresponding pre-trained models or more detailed instructions for reproduction. It seems empty in crop_SIDD.py. Besides, crop_DIV2K.py doesn't match the DIV2K dataset.

Yang-Liu1082 commented 3 years ago

Thank you for your interest in our work. The results reported in the paper are trained on the SIDD dataset with 'LQGTRN_dataset.py'. 'crop_SIDD.py' is only a function to crop the SIDD images into smaller patches. No special preprocessing has been done for that. This file as well as the pre-trained models will be updated soon.