sscnet
sscnet copied to clipboard
have you tested it on ubuntu 16.04
I am compiling with g++ 5.4 under ubuntu 16.04
it seems that the suncg_data_layer is using stdc++11 and this causes the problem of
undefined reference to '*****__cxx11 *****'
I am wondering if you are using 14.04 or 16.04 ? @shurans
I'm using ubuntu 14.04.
successfully compiled caffe_3d on 14.04
@shurans I am trapped by this new bug
cudnn.hpp:145] Check failed: status == CUDNN_STATUS_SUCCESS (3 vs. 0) CUDNN_STATUS_BAD_PARAM
I am using cudnn v5 and cuda-7.5
according to this thread: https://groups.google.com/forum/#!topic/caffe-users/89E0Ci3jjm4
using cuda-8.0 may solve this problem
I have no idea how to hack this, so could you please tell me your cudnn and cuda version ?
If you are using cuda-8.0, then I shall switch to that
I finally got it running.
Using ubuntu 14.04, cudnn-v5 and cuda-8.0.
It seems that using ubuntu 16.04, cudnn-v5.1 or cuda-7.5 will all cause hard-to-hack compatibility bugs.
I have compiled Caffe_3d with ubuntu 16.04 + cuddnn-v5.1 + cuda-8.0.
There exists a small difference in the suncg_data_layer.cu line 273. In g++ 5.4, just let it be:
depth_path = std::string(data_filenames[counter]);
@yuxiaoguo Hi, were you required to update cuDNN.hpp like here?
https://github.com/BVLC/caffe/issues/5793#issuecomment-404729152
@yuxiaoguo Hi, were you required to update cuDNN.hpp like here?
Update: Put aside compiling on Ubuntu 16.04; successfully compiled on Ubuntu 14.04.