Thank you for your outstanding work.
I have two questions that I would like to ask you for your guidance:
(1) Is the training data for the Real_hat_GAN_SRx4.pth model be generated using "Generate degraded images on the fly"?
（https://github.com/xinntao/Real-ESRGAN/blob/master/docs/Training.md#Generate-degraded-images-on-the-fly）
(2) As far as I understand, the dataset used for training is the same as that of real-esrgan (DIV2K + Flickr2K + OST). If we finetune a larger model (HAT-L_SRx4_ImageNet-pretrain.pth) with more data (as used in the SwinIR-L of the swinir project，https://github.com/JingyunLiang/SwinIR), the performance should be better than that of the Real_hat_GAN_SRx4.pth model, right?
Looking forward to your response.
chxy95
commented
1 year ago
Thank you for your outstanding work. I have two questions that I would like to ask you for your guidance: (1) Is the training data for the Real_hat_GAN_SRx4.pth model be generated using "Generate degraded images on the fly"? （https://github.com/xinntao/Real-ESRGAN/blob/master/docs/Training.md#Generate-degraded-images-on-the-fly） (2) As far as I understand, the dataset used for training is the same as that of real-esrgan (DIV2K + Flickr2K + OST). If we finetune a larger model (HAT-L_SRx4_ImageNet-pretrain.pth) with more data (as used in the SwinIR-L of the swinir project，https://github.com/JingyunLiang/SwinIR), the performance should be better than that of the Real_hat_GAN_SRx4.pth model, right? Looking forward to your response.