Something about the test about estrnn

[yuanze-lin]

Hello! I use this code and the parameters in the paper to train on the GOPRO dataset. The results on the test set are lower than the results in the paper. If I want to accurately reproduce the results in the paper, what should I pay attention to?

Hi, What PSNR and SSIM have you achieved?

[lyw3074]

Hello! I use this code and the parameters in the paper to train on the GOPRO dataset. The results on the test set are lower than the results in the paper. If I want to accurately reproduce the results in the paper, what should I pay attention to?

Hi, What PSNR and SSIM have you achieved?

PSNR 29.04 SSIM 0.870

[Iverson-zq]

Hello! I use this code and the parameters in the paper to train on the GOPRO dataset. The results on the test set are lower than the results in the paper. If I want to accurately reproduce the results in the paper, what should I pay attention to?

Hi, What PSNR and SSIM have you achieved?

PSNR 29.04 SSIM 0.870

Hi， I use this code and the parameters in the paper to train on the GOPRO dataset. The LR=e-4, Loss=MSE, n_blocks=9, Channels=16X5=80, batchsize=4 but in the original paper, the batchsize = 8. The results on the test set are also lower than the results in the paper. My results: PSNR =29.1654 SSMI=0.8731. But in the original paper, ESTRNN(B9C80) model‘s results :PSNR = 30.79 ,SSMI=0.9016

[zzh-tech]

The code has been modified based on the previous version.
Try to only change n_blocks=9 and keep the others as default.
Also, it is worth noting that the calculation of metrics in the paper for all models is patch-based.
This version is image-based (the performance for all models will vary accordingly).

[Iverson-zq]

The code has been modified based on the previous version.
Try to only change n_blocks=9 and keep the others as default.
Also, it is worth noting that the calculation of metrics in the paper for all models is patch-based.
This version is image-based (the performance for all models will vary accordingly).

Thank you for your reply. Due to my personal hardware environment is limited (single card 1080Ti, 12G), I set batchsize = 4, and change n_blocks=9, other parameters remain unchanged, the training results: PSNR = 30.72, SSMI = 0.9094. Considering that I reduced batchsize , the result was 0.05 lower than the paper. All in all, thank you very much for your work. I am reproducing the results of your BSD dataset. Thanks again