search

zejinwang / Blind2Unblind

This is an official implementation of the CVPR2022 paper "Blind2Unblind: Self-Supervised Image Denoising with Visible Blind Spots".
https://arxiv.org/abs/2203.06967
120 stars 14 forks source link

Training Results #26

Open wenquanlu opened 1 week ago

wenquanlu commented 1 week ago

Hi, thanks for your great work. I used the command

python train_b2u.py --noisetype gauss25 --data_dir ./data/train/Imagenet_val --val_dirs ./data/validation --save_model_path ../experiments/results --log_name b2u_unet_gauss25_112rf20 --Lambda1 1.0 --Lambda2 2.0 --increase_ratio 20.0

on imagenet validation dataset for training, with all hyperparameters set as default value provided in the README. After 100 epochs, I only get PSNR of 24.75 for PSNR_EXP on Kodak dataset.

What could possibly go wrong?

Thank you

KevinZ0217 commented 1 week ago

Hi, thanks for the great work, I got the same problem during training.