permuting image but not the label

[eliethesaiyan]

@aayushbansal it seems that during seg_image_provider you permute dimension of the image here but i don't see the equivalent in seg_label_provider [here] (https://github.com/aayushbansal/PixelNet/blob/master/experiments/train/utils/seg_label_provider.m) which i beleive might lead to a missmatch of pixels and their annotation and also resizing with different flags(bilinear,antialiasing in seg_image_provider and nearest in seg_label_provider) doesn't have any effect on training?