honorpeter / caffe-deepcompression Goto Github PK
View Code? Open in Web Editor NEWThis project forked from satti007/caffe-deepcompression
Deep Compression follows a pipeline procedure involving Pruning, Quantization and Huffman Coding to compress deep neural network models sizes by 35x-40x. We implemented (pseudo) pruning similar to the first stage of the pipeline on LeNet-5 and decreased it's download bandwidth by 65% (after (tensorlfow) quantization & gzip compression).