Skip to content
This repository has been archived by the owner on Nov 1, 2024. It is now read-only.

Some problem when building with the newest version of caffe2 #60

Open
SunGaofeng opened this issue Dec 17, 2018 · 1 comment
Open

Some problem when building with the newest version of caffe2 #60

SunGaofeng opened this issue Dec 17, 2018 · 1 comment

Comments

@SunGaofeng
Copy link

Thanks for providing us this wonderful repository.
There is some problem when I want to run this code.
(1) When I git clone the code of caffe2 forked by Xiaolong, there is some problem when download submodule codes such as eigen, and lead to a failure of cloning.
(2) Then I try to down load the code of caffe2. Now caffe2 code is merged into pytorch. So I just git clone the pytorch code and build it. As others discussed in the issues, I copy and replace ./caffe2/video to the new code, modify USE_FFMPEG=ON in CMakelist, and fix the bug of inconsistence of protobuf version.
However, building also failed and the error showed that Tensor is not a template. just as the following. I checked the source file of caffe2 forked by Xiaolong, Tensor is actually a Template defined in caffe2/core/tensor.h
template
class Tensor {...}
But in the newest version of pytorch, Tensor in caffe2 is defined in caffe2/core/tensor.h as
class CAFFE2_API Tensor final {...}.
This mismatch will certainly lead to the problem.

============== Error information when build with the newest code of pytorch===========
In file included from /docker_mount/frameworks/pytorch/pytorch/caffe2/video/customized_video_input_op.cc:25:0:
/docker_mount/frameworks/pytorch/pytorch/caffe2/video/customized_video_input_op.h:91:3: error: ‘caffe2::Tensor’ is not a template
Tensor prefetched_clip_on_device_;
^
/docker_mount/frameworks/pytorch/pytorch/caffe2/video/customized_video_input_op.h:92:3: error: ‘caffe2::Tensor’ is not a template
Tensor prefetched_label_on_device_;

@SunGaofeng
Copy link
Author

SunGaofeng commented Dec 17, 2018

So the problem now is that:
(1) submodule can't be cloned when building with the caffe2 code forked by Xiaolong
(2) mismatch of the caffe2_customized_ops/video code with the newest code of caffe2 in pytorch.
Can you supply a complete version of code, including caffe2 as well as thirdparty ?
Or a new version of caffe2_customized_ops/video which has no conflict with the newest version of caffe2?
I've tried several days to rebuilt this excellent model and are really eager to get your help.
Thanks.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant