Skip to content
No description, website, or topics provided.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Instructions to use the code.

This code relies a lot on the projects GauGAN, SRGAN, BasicSR,pix2pix. Credit to these pytorch projects. Some preprocessing code for Tungsten scenes credits to Disney KPCN project.

model weights

experiments dir can be downloaded from googledrive

We provide two set of model weights. "./experiment/kjl/models/opt_XX.pth" is trained from KJL large indoor room datasets without seperating Specular and Diffuse components. You can finetune the weigths to your own dataset based on this pretrained weights.

"./experiments/tungsten_diffuse/models/opt_XX.pth" and "./experiments/tungsten_specular/models/opt_XX.pth" are finetuned on Tungsten scenes.


data dir can be downloaded from googledrive

This directory contain some samples from Tungsten scenes. It also provides the utility scripts to do data processing, for example, to process EXR files, Tungsten data pre-processing ect..

Large scale indoor dataset from will be published soon.


script dir contains json files for training/testing

Attention!!! Change the settings in json files to include your own data path and project root Specifically, remember to change the "dataroot_NOISY" "dataroot_GT" "root" "val_root" to your paths

to run


python -opt script/train/train_seperate_denoiser_diffuse.json
python -opt script/train/train_seperate_denoiser_specular.json


python -opt script/test/test_seperate_denoiser_diffuse.json
python -opt script/test/test_seperate_denoiser_specular.json
You can’t perform that action at this time.