About MSE model about desra HOT 4 CLOSED

For Real-ESRGAN and LDL, we adopt the RealESRNet as the MSE model. Because these two GAN-based models are both fine-tuned from the Real-ESRNet model.

For SwinIR, we trained the MSE model based on the SwinIR setting ourselves, and then added the GAN loss to fine-tune based on this MSE model.

I will provide the models adopted in the paper.

from desra.

对于Real-ESRGAN和LDL，我们采用RealESRNet作为MSE模型。因为这两个基于 GAN 的模型都是从 Real-ESRNet 模型微调而来的。

对于 SwinIR，我们自己根据 SwinIR 设置训练了 MSE 模型，然后添加了 GAN 损失以基于此 MSE 模型进行微调。

我将提供本文中采用的模型。

Thank you very much! When load the checkpoints in mmseg0.29.0,I have got this problem"model.CLASSES = checkpoint['meta']['CLASSES']
KeyError: 'meta'".This checkpoints is true？

from desra.

I just uploaded the checkpoints to Google Drive, you can get them from this link (https://drive.google.com/drive/folders/10vzKAJjHm5GEb0_LESnHYmuCnBwdoFpm). The corresponding information has also been updated on the project homepage.

from desra.

For Real-ESRGAN and LDL, we adopt the RealESRNet as the MSE model. Because these two GAN-based models are both fine-tuned from the Real-ESRNet model.

For SwinIR, we trained the MSE model based on the SwinIR setting ourselves, and then added the GAN loss to fine-tune based on this MSE model.

I will provide the models adopted in the paper.

有一点我不太理解，RealESRGAN的论文里写Real-ESRNet用的是L1 LOSS，为什么这里说他是MSE（L2 LOSS）的模型呢？谢谢！

from desra.