To use the Computer Vision functions with a Pylearn2 CNN and also with a hierarchical, stacked model, it would be much faster if we can cache the CV outputs so they don't have to be recomputed.
I think we should take the hash of the HLF object and the hash of the image and use these to store/load the output feature vector.
To use the Computer Vision functions with a Pylearn2 CNN and also with a hierarchical, stacked model, it would be much faster if we can cache the CV outputs so they don't have to be recomputed.
I think we should take the hash of the HLF object and the hash of the image and use these to store/load the output feature vector.