Open essalim opened 9 years ago
maybe u can try this: for a single channel image, just copy this channel three times and make a three channel image
The problem is not the number of channels, i am just thinking that the normalization is application dependant.
Hi, I am working on grayscale face images, all the images are just a face on a white/gray background, the mean normalization doesn't seem to work well with a single object on a background images,what would be a better pre-processing for this kind of images befor network training?