MobileNet implementation for torch based on Pytorch Equivalent: https://github.com/marvis/pytorch-mobilenet
This is a Torch implementation of Mobilenet architecture.
a. Add bias once the nn library solves the issue with Bias on DepthWiseSeparable convolution.
b. Add training on imagenet and food101.
c. Add pretrained model and demo
Training is very slow. In particular, it takes about 300msec to process a batch of 10 images (30msec/image) on a NVIDIA GTX-1070.
Please feel free to contact me for any comments or suggestions.