The fundamental difference between regular Convolution Networks to MobileNets architecture is that it takes hugely less number of parameters to train and by doing so more efficient than the regular CNNs. Efficiency has a trade-off with accuracy: would that architecture fit to your use case?
Paper: Visual Search at Pinterest by Jing et al. Visual Discovery, Pinterest and University of California, Berkeley.
About Me
In two words: lifelong learner. I’m a product management leader with my expertise spanning over two decades. With a unique blend of product and technology, I help deliver industry-grade solutions from concept to cash (and beyond). more about me