MonocularTotalCapture
MonocularTotalCapture copied to clipboard
Dockerfile or Docker image Requirement
Thanks for your great work! I find it difficult to configure the environment, would it be possible for you to release a dockerfile or docker image? Thanks in advance~
@AlexHu123 Have you solve the fussy environmental work? a docker image would be helpful
Thanks for your reply! I have solved the environmental work. It is quite tricky and needs to be careful at every step~ If having time, I may write a blog on how to install it, and will let you know as reference~
@AlexHu123 Hi,I meet some problem:
Root Directory: /home/nd/MonocularTotalCapture/data
Sequence Name: example_dance
run_pipeline.sh: line 58: 16476 Segmentation fault (core dumped)
./build/run_fitting --root_dirs $dataDir --seqName $seqName --start 1 --end $((numFrame +1)) --stage 1 --imageOF
Detail in issue 33#
Have you also meet the problem before? Any help would be greatly appreciated.
See PR https://github.com/CMU-Perceptual-Computing-Lab/MonocularTotalCapture/pull/35 for the docker image
@neccam I'm building your image and I keep running into Caffe not found. Either turn on the BUILD_CAFFE option or specify the path of Caffe includes and libs using -DCaffe_INCLUDE_DIRS and -DCaffe_LIBS
.
This happens when building OpenPose :
cmake -DCaffe_INCLUDE_DIRS=$CAFFE_ROOT/build/install/include \ -DCaffe_LIBS=$CAFFE_ROOT/build/install/lib/libcaffe.so \ -DBUILD_CAFFE=OFF \ -DDOWNLOAD_BODY_COCO_MODEL=ON \ -DDOWNLOAD_BODY_MPI_MODEL=ON \ .. &&
DCaffe_INCLUDE_DIRS=$CAFFE_ROOT/build/install/include
and -
DCaffe_LIBS=$CAFFE_ROOT/build/install/lib/libcaffe.so` don't seem to be recognized by cmake.
What do I do?
It should work as is, but I will have a look into it. In the meantime, it would be good to know if anyone else has a similar issue.
This seems to be an issue with OpenPose. Have a look at this issue : https://github.com/CMU-Perceptual-Computing-Lab/openpose/issues/1575
It's an issue with building openpose with pre-built caffe. Is there a reason you build caffe first and then openpose? You seem to make a few changes to the caffe makefile in the dockerfile.
Update. Solved it by using the previous OpenPose release. @neccam Does the visualization work if I'm running the container on a remote server?
Great, can you share what you have changed so that others can also benefit?
Depends on how you access to the remote server. My best guess would be to setup a ssh deamon inside the docker container and ssh -x to it. But I haven't tried this.
Couple of changes:
-
git reset --hard d78ae77
OpenPose commit on Jan 20th. Assumed this would be the one you used - Add these flags.
-DCUDA_ARCH=Manual -DCUDA_ARCH_BIN="50 52 60 61" -DCUDA_ARCH_PTX="50 52 60 61"
. Cuda 9.0 doesn't work on a GPU with compute capability of 7.5 (T4, RTX series) OpenPose builds for all cc's and the build fails. So manually specify.
Here's the code block. Copy and replace in DockerFile.
ENV OPENPOSE_ROOT=/opt/openpose WORKDIR $OPENPOSE_ROOT RUN git clone https://github.com/CMU-Perceptual-Computing-Lab/openpose.git . && \ git reset --hard d78ae77 && \ sed -i 's/set(Caffe_known_gpu_archs "${KEPLER} ${MAXWELL} ${PASCAL} ${VOLTA} ${TURING}")/set(Caffe_known_gpu_archs "${KEPLER} ${MAXWELL} ${PASCAL}")/' /opt/openpose/cmake/Cuda.cmake && \ mkdir build && cd build && \ cmake -DCaffe_LIBS=$CAFFE_ROOT/build/install/lib/libcaffe.so \ -DCaffe_INCLUDE_DIRS=$CAFFE_ROOT/build/install/include \ -DBUILD_CAFFE=OFF \ -DDOWNLOAD_BODY_COCO_MODEL=ON \ -DDOWNLOAD_BODY_MPI_MODEL=ON \ -DCUDA_ARCH=Manual -DCUDA_ARCH_BIN="50 52 60 61" -DCUDA_ARCH_PTX="50 52 60 61" .. && \ make all -j8 && \ make install