Training Imagenet using Mocha

[vollmersj]

Has anyone trained imagenet or parts of imagenet using Mocha or just loaded parts of the data set for backpropgation. Any recommendations about the setup would be greatly appreciated.

[pluskid]

@vollmersj I personally have not done that. For training with huge ImageNet dataset, I believe converting the data into HDF5 format is very infeasible. Actually, IO might become the bottleneck at some point. Caffe use specific LevelDB or LMDB to store the images to get high speed random access. I think to make this work, a first thing needed is to at least to write a specific DataLayer to load from proper format of the dataset.