Could you please specify the usage of your code? THX

[JustinAsdz]

Hi Elin

Thanks for your work, recently I want to reproduce the result of your method. Can you specify the work-flow of your paper, for convenient usage for others to use.

Thanks !

[Elin24]

There are 6 steps for an input LR image and its ground truth HR:

1. use a super-resolution network to generate SR image;
2. figure out the SSIM map between HR and SR, corresponding to the following code:

   ssim = pytorch_ssim.ssim(hr, sr, reduction='none').detach()

3. use the gaussian function to generate attention map:

   # the meaning of sigma and maxVal can be found in my paper
   gauss = lambda x: torch.exp(-((x+1) / sigma) ** 2) * self.maxVal
   weight = gauss(ssim).detach()

4. generate new HR and new SR image:

   nsr = sr * weight
   nhr = hr * weight

5. adopt loss function like mse or mae to calculate final loss value:

   loss = mae(nsr, nhr) # use mae as example

6. back propagation, which can be done by Pytorch use loss.backward()

---

PSPL is a very sample method to accelerate convergence, all its function can be represented by the code show in Core Part in README file.

[JustinAsdz]

Thanks for your quick reply,

Your Explanation is clear and percise, but can you show the steps to run the code for example : about dataset preparation or running file order. Thanks

[Elin24]

I have explained it in issue 1, including preparing data and running the code in src. You can refer to it.

[JustinAsdz]

Thanks ,go to issue 1