Manage Okutama dataset: splits, merge?, update files, structure

[miquelmarti]

At the moment I was assuming that each split should have a subfolder with images, ground truth annotation, color scheme etc. Not even there, no splits yet for Hikawa and Okutama.

Would be nice to change this so all the images etc. reside under same folder, just have different lists for each split. Kind of what it was done originally but each dataset has its own lists. See #3 .

• Should we merge Hikawa and Okutama?
• Crop images of Okutama2 dataset?
• Create splits
• Data normalization (channel mean subtraction. Scaling and mean image look like are not normally done in caffe community)
• Update READMEs

[miquelmarti]

Also substract mean! There is a script around I think