Optimizing Conv in Caffe
- Solution: converting Conv to GEMM
This is done in two lasy steps:
(1) do an im2col to convert the image to a matrix
(2) call GEMM to do the actual computation - Pros:
simple logic, easy implementation -
Cons:
Lots of memory overhead
11.jpg
12.jpg
13.jpg
14.jpg
15.jpg
参考论文：
High Performance Convolutional Neural Networks for Document Processing
