A user asked us to install a software named openpose on Baobab. This software needs a lot of dependencies and we didn’t had enough time to install it. More an more software are available as Docker image as well. That’s great because we support Docker like images through Singularity.
Steps:
build the image for singularity
launch the container
According to this page, the Docker image of openpose is available as garyfeng/docker-openpose.
Let’s build the Singularity image openpose. This will download the Docker images and takes some time. It will also produce some warnings because we are running this command as normal user instead as user root.
Thanks for the tutorial.
A very useful singularity flag is the -B flag which allows to mount an external folder to a folder inside the container. This is very useful when manipulating huge data file which should be on scratch and not in the home directory.
For example the flag
-B /home/strassp6/scratch/Data:/Data
will mount the Data folder on scratch to /Data.
To add more information about what Yann said. A good source of docker images is dockerhub and especially for gpu [ngc nvidia](https://ngc.nvidia.com/].
Hello,
I’m trying to use openpose through a docker either with " aryfeng/docker-openpose:latest" or " tchataing/op:latest" but i’m confronted with a CUDA error.
Error:
Cuda check failed (209 vs. 0): no kernel image is available for execution on the device
I pulled the docker image through singularity build docker://... or singularity pull docker://...
and tried to run it but I can’t manage the CUDA version properly.
Do you have any insight on how CUDA should be handle between the container and the job ?
#!/bin/sh
#SBATCH --partition=shared-gpu
#SBATCH --gpus=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=1
#SBATCH --time=10:00
## https://hub.docker.com/layers/sandeepagowda/openpose/CUDA12.0/images/sha256-7cbc77f912e0b9e7d3035e6e5bc010d34e56a88f7e6a1a1ecca24963afa6cf01
OPENPOSE_URL=docker://sandeepagowda/openpose:CUDA12.0
OPENPOSE_SIMG=openpose.simg
VIDEO_INPUT=/home/share/schaer2/thibaut/opensam/input/balt.mp4
VIDEO_OUTPUT=/home/share/schaer2/thibaut/opensam/output
srun apptainer build $OPENPOSE_SIMG $OPENPOSE_URL
## not all the flags are working with this newer version
srun apptainer exec --nv --pwd /openpose $OPENPOSE_SIMG build/examples/openpose/openpose.bin --video $VIDEO_INPUT --write_video $VIDEO_OUTPUT --display 0
By hand with :
(baobab)-[chataint@gpu008 thibaut]$ apptainer build openpose.simg docker://sandeepagowda/openpose:CUDA12.0
(baobab)-[chataint@gpu008 thibaut]$ apptainer run --nv openpose.simg
Apptainer> cd /openpose/
Apptainer> build/examples/openpose/openpose.bin --display 0 --video /home/share/schaer2/thibaut/opensam/input/balt.mp4 --write_video /home/share/schaer2/thibaut/opensam/output
Starting OpenPose demo...
Configuring OpenPose...
Starting thread(s)...
Auto-detecting all available GPUs... Detected 1 GPU(s), using 1 of them starting at GPU 0.
Error:
Cuda check failed (209 vs. 0): no kernel image is available for execution on the device
Coming from:
- /openpose/src/openpose/net/netCaffe.cpp:reshapeNetCaffe():115
- /openpose/src/openpose/gpu/cuda.cpp:cudaCheck():37
- /openpose/src/openpose/net/netCaffe.cpp:reshapeNetCaffe():120
- /openpose/src/openpose/net/netCaffe.cpp:forwardPass():259
- /openpose/src/openpose/pose/poseExtractorCaffe.cpp:forwardPass():632
- /openpose/src/openpose/pose/poseExtractor.cpp:forwardPass():53
- /openpose/include/openpose/pose/wPoseExtractor.hpp:work():107
- /openpose/include/openpose/thread/worker.hpp:checkAndWork():93
Error occurred on a thread. OpenPose closed all its threads and then propagated the error to the main thread. Error description:
Cuda check failed (209 vs. 0): no kernel image is available for execution on the device
Coming from:
- /openpose/src/openpose/net/netCaffe.cpp:reshapeNetCaffe():115
- /openpose/src/openpose/gpu/cuda.cpp:cudaCheck():37
- /openpose/src/openpose/net/netCaffe.cpp:reshapeNetCaffe():120
- /openpose/src/openpose/net/netCaffe.cpp:forwardPass():259
- /openpose/src/openpose/pose/poseExtractorCaffe.cpp:forwardPass():632
- /openpose/src/openpose/pose/poseExtractor.cpp:forwardPass():53
- /openpose/include/openpose/pose/wPoseExtractor.hpp:work():107
- /openpose/include/openpose/thread/worker.hpp:checkAndWork():93
- [All threads closed and control returned to main thread]
- /openpose/src/openpose/utilities/errorAndLog.cpp:checkWorkerErrors():280
- /openpose/include/openpose/thread/threadManager.hpp:stop():243
- /openpose/include/openpose/thread/threadManager.hpp:exec():202
- /openpose/include/openpose/wrapper/wrapper.hpp:exec():424
Feel free to use the same video “balt.mp4” for test if needed.
not an official release)