medicinestick / caffemodel_compress

A c++ project what performing CNN channel pruning

Caffemodel_Compress# <<Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding>>

Song Han, Huizi Mao, William J. Dally (Submitted on 1 Oct 2015 (v1), last revised 15 Feb 2016 (this version, v5))

A c++ project what performing CNN channel pruning It's an idea I've had floating around for a while , You can perform channel pruning on your trained caffemodel by using this tool, A XML file you should recollocate first,irrespective of whether depthwise or convolution layer, when convolution layer which need to prune has been setting, a reciprocal arrangement of channel pruning also has been setting during processing of progress implicitly, then do a little path change in main.cpp ,and build it with your local IDE.

updated 18/6/25 Already supported to eltwise pruning mac...

caffemodel_compress's Issues

can you explain how your method works?

Hi, lusenkong
I am very glad that you share your code with us, can you explain how u algorithm works？
can u share some demo about how to use Caffemodel_Compress?

请教一些疑问

非常感谢博主的开源代码。对于您的算法有点小疑问：

1、你的算法是对训练完成后的caffemodel和prototxt进行模型压缩，那如果这个caffemodel本身精度还没有训练到收敛，是否可以把Compress后的prototxt重新训练(不是findtune)一个新的压缩版caffemodel?
2、还是说原始的caffemodel就要训练得很好，然后再通过您的代码进行模型压缩得到新的caffemodel和prototxt，并且通过微调训练之后得到最终的部署caffemodel?
3、压缩之后的模型精度是否会大致按比例损失吗？

望博主解答一二，谢谢！

配置好caffe为什么还是跑不了你的代码。

你好，我在Windows上编译好了caffe。但是还是跑不了你的代码，请问一下还需要配置什么东西吗？

关于全连接层

您好，我在学习您的源码，还没有测试。
没有看到对于全连接层的处理，是否无法修剪卷积层+全连接层的模型，
感谢您的查看

你好，你的代码主要依是那篇文章做的剪枝？

请问对mobilenet-ssd同样可以用吗