about data process

Lines 31 to 35 in 685d4af

    
           elif dataset == 'koniq-10k': 
        
               if istrain: 
        
                   transforms = torchvision.transforms.Compose([ 
        
                       torchvision.transforms.RandomHorizontalFlip(), 
        
                       torchvision.transforms.Resize((512, 384)),

hyperIQA/folders.py

Line 191 in 685d4af

sample.append((os.path.join(root, '1024x768', imgname[item]), mos_all[item]))

From the two code snippets above, it is evident that the data is loaded as images with a size of '1024x768', with width and height dimensions of 1024 and 768 respectively. However, the Resize((512, 384)) operation rescales the dimensions to 512 and 384, resulting in a noticeable change in the aspect ratio from the original 768:1024 to 512:384. I'm curious if the same processing is applied in the experimental setup of the paper?

Plus: according to the document of pytorch Resize,the 'size' parameter of the Resize function refers to the height and width.

	elif dataset == 'koniq-10k':
	if istrain:
	transforms = torchvision.transforms.Compose([
	torchvision.transforms.RandomHorizontalFlip(),
	torchvision.transforms.Resize((512, 384)),