Can not re-produce SR result on DIV2K dataset

[Arcananana]

Thanks for your work. When I try to re-produce the SR results using the pre-trained models, the reconstructed images contains lots of noises like this {'psnr': 14.266, 'ssim': 0.115, 'lpips': 1.334, 'lr_psnr': 25.924} total: 1

[2590609710]

Thanks for your work. When I try to re-produce the SR results using the pre-trained models, the reconstructed images contains lots of noises like this {'psnr': 14.266, 'ssim': 0.115, 'lpips': 1.334, 'lr_psnr': 25.924} total: 1

Hello, I also encountered the same problem. May I ask how did you solve this issue?

[Ecnu-luobote]

The pre-model file should be in the checkpoints/srdiff_pretrained_div2k folder

[duanpanjun]

The pre-model file should be in the checkpoints/srdiff_pretrained_div2k folder预模型文件应位于 checkpoints/srdiff_pretrained_div2k 文件夹中

This is what I did, but the results still contain a lot of noise. I don't understand why this is happening.

[iF1844680954]

I also encountered the same problem,Could you please tell me how you solved it？

[iF1844680954]

I also encountered the same problem,Could you please tell me how you solved it？

It seems I've solved this issue. I made the "binary_data_dir" in" configs/config_base.yaml" and "configs/df2k4x.yaml" consistent, setting the path to "binary_data_dir: data/binary/df2k4x", and then reran python :"data_gen/df2k.py". When testing, please use the terminal command:" python tasks/trainer.py --config configs/diffsr_df2k4x.yaml --exp_name srdiff_pretrained_div2k --infer."