Apollo: 交通信号灯识别模块caffe net的输入与输出层问题

Created on 22 Jan 2018  ·  13Comments  ·  Source: ApolloAuto/apollo

apollo-2.0.0\modules\perception\traffic_light\rectify\detection.h中

include "caffe/layers/pyramid_layers.hpp"

安装编译caffe后,在caffe/layers中找不到/pyramid_layers.hpp
请问,pyramid_layers.hpp是否是apollo自己定义的,在哪里能够找到呢

Perception Question

All 13 comments

Are you running the program in docker?

我没有在docker中运行
我只是把detect这部分代码单独拿出来,自己新建了一个工程,cmake后make

@yongyuandeyezi 据我所知,caffe这部分现在是不开源的。

for your info:
Some libraries are only inside docker. https://github.com/ApolloAuto/apollo/issues/1751

My suggestion is that if you would like to play around without docker environment as a developer, you can build your own network layers.

For pyramid networks, here is a paper for your reference (this might be different from what perception team is using, but it could provide some related information to help you build your own one).
Feature Pyramid Networks for Object Detection
https://arxiv.org/abs/1612.03144

caffe in docker is different from origin version and is not open source in Apollo 2.0. If you want to use outside docker please copy caffe's *.h and *.so to your environment. Or wait for Apollo 2.5

Thank you very much!

seemed still no open source plan for traffic light module in Apollo 2.5 @ghdawn

Are all neural networks not open sourced ?

The caffe code is all open source now.

@techoe where is caffe code that is open source in apollo?

Hi @breakisx , the code is in another repo: https://github.com/ApolloAuto/caffe
We build it and put the output into Apollo docker container: https://github.com/ApolloAuto/apollo/blob/master/docker/build/installers/install_gpu_caffe.sh

Hi @breakisx , the code is in another repo: https://github.com/ApolloAuto/caffe
We build it and put the output into Apollo docker container: https://github.com/ApolloAuto/apollo/blob/master/docker/build/installers/install_gpu_caffe.sh

Hi,the pyramid_layers.hpp couldn‘t be found in the apollo/caffe.

@xiaoxq the caffe code link is not available now.

Was this page helpful?
0 / 5 - 0 ratings

Related issues

chilihua picture chilihua  ·  3Comments

lyletbjohnson picture lyletbjohnson  ·  3Comments

JSnobody picture JSnobody  ·  3Comments

BenBaek picture BenBaek  ·  3Comments

Wsine picture Wsine  ·  3Comments