Same as the original version of Caffe.
See http://caffe.berkeleyvision.org/installation.html for the latest installation instructions.
Two steps are required to prune a neural network. Suppose we have a pretrained model "bvlc_alexnet.caffemodel".
Step 1: admm (suppress target weights)
./build/tools/caffe train --solver=models/bvlc_alexnet/solver_admm.prototxt --weights=/path/to/bvlc_alexnet.caffemodel
Step 2: retrain (set target weights to 0 and retrain)
./build/tools/caffe train --solver=models/bvlc_alexnet/solver_retrain.prototxt --weights=models/bvlc_alexnet/caffe_alexnet_train_admm_iter_2400000.caffemodel
Suppose caffe_alexnet_train_admm_iter_2400000.caffemodel is the trained caffemodel from step 1.
- lr_policy: "admm" for admm step; your prefered policy for retrain step
- gamma: drop the learning rate by a factor of this value
- stepvalue: drop the learning rate by gamma when meet this stepvalue in each admm iteration
- admm_iter: Caffe training iterations for each admm iteration
- max_iter: Total training iterations (total admm iterations = max_iter / admm_iter)
- pruning_phase: "admm" for admm step and "retrain" for retrain step
- prune_ratio: the percent of weight parameters to prune for each layer
- rho: the rho parameter for admm pruning (detailed in our paper)
Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by Berkeley AI Research (BAIR)/The Berkeley Vision and Learning Center (BVLC) and community contributors.
Check out the project site for all the details like
- DIY Deep Learning for Vision with Caffe
- Tutorial Documentation
- BAIR reference models and the community model zoo
- Installation instructions
and step-by-step examples.
- Intel Caffe (Optimized for CPU and support for multi-node), in particular Xeon processors (HSW, BDW, SKX, Xeon Phi).
- OpenCL Caffe e.g. for AMD or Intel devices.
- Windows Caffe
Please join the caffe-users group or gitter chat to ask questions and talk about methods and models. Framework development discussions and thorough bug reports are collected on Issues.
Happy brewing!
Caffe is released under the BSD 2-Clause license. The BAIR/BVLC reference models are released for unrestricted use.
Please cite Caffe in your publications if it helps your research:
@article{jia2014caffe,
Author = {Jia, Yangqing and Shelhamer, Evan and Donahue, Jeff and Karayev, Sergey and Long, Jonathan and Girshick, Ross and Guadarrama, Sergio and Darrell, Trevor},
Journal = {arXiv preprint arXiv:1408.5093},
Title = {Caffe: Convolutional Architecture for Fast Feature Embedding},
Year = {2014}
}