search

hjimce / compress_yolo

对yolo模型进行网络压缩、修剪等实验
Other
50 stars 33 forks source link

same model size #1

Open fishman2008 opened 7 years ago

fishman2008 commented 7 years ago

Do you need to change the setting in CFG file to prune the model? Setting "prune=1" in makefile only is not helping the model reduction. What else changes do I have to make? Thanks.

hjimce commented 7 years ago

@fishman2008 I only commit the network pruning code，you have to use Huffman Coding compress it by youself。

csaimd commented 7 years ago

你好，hjimce,我是一名研究生菜鸟，导师给的方向是yolo的压缩轻量化，能否给个联系方式，邮箱什么的，跟你请教一下

hjimce commented 7 years ago

@csaimd 你可以看看网络优化相关的文献，先学学网络优化

hvudeshi commented 5 years ago

Hello @hjimce,

I have applied your pruning process in my cfg and pre-trained weights, but I haven't seen any reduction in weight file. But, here you mentioned about Huffman Coding, so just pruning won't help? Also, can you guide me regarding the implementation details of Huffman Coding?

sicarioakki commented 5 years ago

@hjimce Hi man! I have also tried implementing your code, but i donot see any reduction in weights file size. @hardik22317 did you try Huffman coding?

hvudeshi commented 5 years ago

@hjimce Hi man! I have also tried implementing your code, but i donot see any reduction in weights file size. @hardik22317 did you try Huffman coding?

I haven't tried yet, but I am finding ways to implement it and the exact position, I mean the function of darknet in which I could implement the Huffman Coding. I am not able to understand how to consider the frequencies of weight for Huffman Coding. Any help is always welcomed.