1

I have almost 5223 and 577 images for training and validation sets, respectively. I am applying CNNs for image segmentation and would like to do artificial on-the-fly data augmentation. I have installed Caffe latest version. I am wondering whether this version of Caffe is supporting data augmentation or not? If yes, could you please share some resources for me? The other question is whenever we are doing artificial data augmentation, should we change the epoch according to the augmentation size? for example, if I only apply mirroring, should I change the size of epoch by mutiplying 2?

S.EB
  • 1,966
  • 4
  • 29
  • 54
  • 1
    i think rotating image and varying pixel intensity(dividing images pixel values by a number smaller than 255) could do,i did the same on on segnet but i didnt improve the quality,pixel level segmentation with the image you have is enough i think since the features are learned on pixel level rather than the whole image,therefore you have enough features if those image are all different but you might consider dropping out where your model accuracy stagnates – Eliethesaiyan May 22 '17 at 11:50
  • @Eliethesaiyan Thank you very much for your guidance. I was thinking it is not enough for training. – S.EB May 22 '17 at 11:55
  • 1
    i think in general,pixel level segmentation doesn't need much image,if you look at Camvid,they have only used 400 and got better results...i think where it matter the most is object recognition where object can take different forms as the object must be recognized as whole whereas in segmentation...each pixel is matched with the corresponding pixel label...here patterns don't count so much in my opinion... – Eliethesaiyan May 22 '17 at 11:59
  • @Eliethesaiyan Thank you very much once again. – S.EB May 22 '17 at 12:00

0 Answers0