search

BVLC / caffe

Caffe: a fast open framework for deep learning.
http://caffe.berkeleyvision.org/
Other
34.06k stars 18.7k forks source link

Deep compression with Caffe #4789

Open naranjuelo opened 8 years ago

naranjuelo commented 8 years ago

Hi! I was wondering if there is any possibility in Caffe to compress a neural network.

In this paper Deep compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding , they could reduce for example the size of VGG-16 by 49x without affecting its accuracy (very interesting in networks that require so much memory).

Any way to remove the connections with weights below a threshold? Thank you very much!

sivagnanamn commented 7 years ago

We've tried similar thing by writing a wrapper in C to access the .caffemodel weights and it worked. You can try your own custom wrapper function for pruning.