about input

[lichunxiang93]

Thank you very much for sharing, I have some questions here.Is the input of DIQA also patch? Can the whole picture be input directly? I did not see the details of patch input in the paper.

[lllllllllllll-llll]

Hello! For the training step, the input of the network is the image patch (e.g., 112x112 pixels patch for LIVE dataset). For the training details, you can refer to the describes of the original paper "G. Patch-Based Training", "In the DIQA framework, the sizes of input images must be fixed to train the model on a GPU. Therefore, to train the DIQA using images of various sizes, such as in the LIVE IQA database [5], each input image should be divided into multiple patches of the same size. Here, the step of the sliding window is determined by the patch size and the number of ignored pixels around the borders to avoid overlapping regions when the perceptual error map is reconstructed."

[Sere1nz]

Can we resize the input image using transform function? or we must divide it into patches

[Sere1nz]

Can we resize the input image using transform function? or we must divide it into patches

I might also change the input size of the model if I am using a dataset that input size is not that large and the same. In this case I don't need patch-based training. Is that right?