Parallel Decoder

Parallel decoder is a data streaming source which works like native Python generator, except that the computational part of generating data sample is delegated to a multi-processing computing module. It can be treated as a parallel version of generator.

Motivation

The idea of this project came from my last machine learning project which is about human face comparison. We were training a Siamese network which takes a pair of images as input each time. As we increase complexity of image decoding, the data feeding speed becomes the bottlenect, so I came up with this idea.

A Parallel Decoder is designed to have following features: