Open Vadim2S opened 3 years ago
Are you talking about training? In training, auth and tamp are not used as pairs. They are used independently. You can use any auth and tamp images. You may even use only tamp images and no auth images.
Big thanks. Another questions: 1) Is tampering and groundtruth files must be same shape? You check it and drop mismatched files in CASIA adapter only. Other datasets adapters do not check shape. 2) You provide auth files as single file and tampered files as tampered file and groundtruth file or None. How much importance of groundtruth file here? I am can remove tampered file without groundtruth pair if it worth to do so.
Sorry for my terrible english.
1) Where gt is resized? Must i am resize gt myself or you CAT-Net code do it? P.S. In dataset_CASIA.py you just discard misshaped tp+gt pair and do not use them at all. 2) This is stupid question about using tp without corresponding gt, sorry.
Is they must be same size, aspect ratio and compression? Is images may be slightly moved relative center of picture?