search

yulunzhang / RCAN

PyTorch code for our ECCV 2018 paper "Image Super-Resolution Using Very Deep Residual Channel Attention Networks"
1.39k stars 316 forks source link

Question about the number of training set #72

Open bberry25 opened 5 years ago

bberry25 commented 5 years ago

Hi, I have a question about the number of the train set. After the augmentation process of the training set in your code, How many times do the data increase after that process is over?

yulunzhang commented 5 years ago

Hi, Data augmentation is conducted online. The physical data is not augmented, but the network experienced training data about 8 times to the original data without augmentation.

bberry25 commented 5 years ago

Thank you for your reply. But I'm a little bit confused about the training process. I read that you did data augmentation on the 800 training images which are randomly rotated by 90, 180, 270 and flipped horizontally in your journal and found augmentation process in common.py. If the physical data is not augmented, how the network experienced training data about 8 times to the original data without augmentation? Where can I check this process in your code or could you explain more specific explanation information about train process? Thanks.

qingchuanhuajuan commented 5 years ago

@yulunzhang Hello, I want to turn off data augmentation because my data is big enough. what should I do? Is the ‘def augment(l, hflip=True, rot=True):’ in the common.py file changed to ‘def augment(l, hflip=False, rot=False)’?