CaffeConTroll on mobile devices

[rhythm9229]

Will the techniques for speeding up neural networks used in CaffeConTroll (lowering techniques and batching) work well for mobile devices?

[blueardour]

In my understanding, the lowering techniques is similar to the im2col function in the caffe.

And the batching is to set a large batch size when training, distributed to CPUs and GPUs dynamically according to their FLOPS. For example, 32 images in one batch, CPUs have 1 GOPS ability and GPUs have 3GOPS, then 32*(1/4) on CPUs, 32*(3/4) on GPUs.

If my understanding is correct, I think the answer to your question is Yes. Just dispatch different ratio of input data in one batch to mobile CPU, GPU or DSP based on the processing ability.