Jie Zhang
Jie Zhang
这个代码确实写错了,应该是 idx_part 😂
Yes. The cascade framework can be used to detection objects with fixed window size.
memory pool manager is used for every thread, and these instances only freed when the program exit. check the code [here](https://github.com/luoyetx/mini-caffe/blob/master/src/thread_local.hpp#L34) and [here](https://github.com/luoyetx/mini-caffe/blob/master/src/syncedmem.cpp#L142)
你是怎么监测到 900 M 的内存占用的,下面的图是我在 Debug 模式下跑 SSD exmaple 画出来的  [mem.log](https://github.com/luoyetx/mini-caffe/files/1953424/mem.log)
@wait1988 Linux 崩溃似乎是 protobuf 的问题?
我本地 debug 模式下跑的 ssd example 
3.4 版本,内存占用还是有问题吗?
All reshape stuffs are done in [PlaceMemory](https://github.com/luoyetx/mini-caffe/blob/master/src/net.cpp#L147). When the shape of input blob changes, all internal blob will change the shape and realloc the memory buffer.
The layer itself gets all shape info about input blobs, it should be able to computer the shape of output blobs when reshape function called.
For proposal layer, checkout code [here](https://github.com/luoyetx/mini-caffe/blob/master/src/layers/proposal_layer.cpp#L249). We set maximum shape for output rois.