About CameraSR dataset about dasr

[House-Leo]

And how do you get the test set(iphoneX subset) LR images? Bicubic downsample?
In the CameraSR dataset https://github.com/ngchc/CameraSR, LR has the same size as HR

[DizzyYunxuan]

Thanks for your attention.

1.We train our DASR by DIV2K and only 21-100 LR images in CameraSR iphoneX subset, and test on 0-20 image pairs. We did not use Nikon data.

2.As mentioned in the paper that "As the LR and HR images in the dataset are with the same spatial size, we remove the down-sampling and up-sampling operations in our framework as well as the FSSR model." in section 5.2.

[DizzyYunxuan]

2.As mentioned in the paper that "As the LR and HR images in the dataset are with the same spatial size, we remove the down-sampling and up-sampling operations in our framework as well as the FSSR model." in section 5.2.

It means that the input LR and output HR are of the same spatial size.

[House-Leo]

Thanks for your attention.

1.We train our DASR by DIV2K and only 21-100 LR images in CameraSR iphoneX subset, and test on 0-20 image pairs. We did not use Nikon data.

2.As mentioned in the paper that "As the LR and HR images in the dataset are with the same spatial size, we remove the down-sampling and up-sampling operations in our framework as well as the FSSR model." in section 5.2.

Thanks for your answer! It help me a lot!

[House-Leo]

2.As mentioned in the paper that "As the LR and HR images in the dataset are with the same spatial size, we remove the down-sampling and up-sampling operations in our framework as well as the FSSR model." in section 5.2.

It means that the input LR and output HR are of the same spatial size.

Did it mean that you use 21-100 LR(in the iphoneX subset with the same spatial size as HR) and DIV2K_HR in the training phase, and test on 0-20 LR(input)?

[DizzyYunxuan]

Did it mean that you use 21-100 LR(in the iphoneX subset with the same spatial size as HR) and DIV2K_HR in the training phase, and test on 0-20 LR(input)?

Yes.

[House-Leo]

Did it mean that you use 21-100 LR(in the iphoneX subset with the same spatial size as HR) and DIV2K_HR in the training phase, and test on 0-20 LR(input)?

Yes.

Hi, Yunxuan!
Can you provide the test option (json or yml file) for CameraSR dataset? I'm sorry that I can't find it in options/test/test_sr.json.