A tool will be implemented to operate a surgery to "sparsify" caffemodels according to config files. The source code can be found in this repo.
Recently, many works about pruning CNN to accelerate inference have been published, such as Net-Trim, NISP and ThiNet. The being implemented tool can be used to operate the surgery of pruning CNN based on Caffe toolkit.