I created a Dockerfile that uses the latest OpenPose version (rel https://github.com/fabiozappo/SkeletonGroupActivityRecognition/issues/4). Previously, there has been two containers (one for OpenPose an one for Skeleton Group Activity Recognition). Now, there has only one container to be created that has the ability for both skeleton extraction and group activity recognition. (Therefore I have also removed Docker Compose.)
Currently, nvidia/cuda:10.0-cudnn7-devel is used as base image. I also tried docker pull nvidia/cuda:11.0.3-cudnn8-devel (what can be built as well), but that will run into a cudaSuccess (2 vs. 0) out of memory error - according to https://github.com/CMU-Perceptual-Computing-Lab/openpose/issues/1230#issuecomment-774706454 newer cuDNN versions seem to need more memory for the same model.
The container inside which everything works comes with the following details:
I created a Dockerfile that uses the latest OpenPose version (rel https://github.com/fabiozappo/SkeletonGroupActivityRecognition/issues/4). Previously, there has been two containers (one for OpenPose an one for Skeleton Group Activity Recognition). Now, there has only one container to be created that has the ability for both skeleton extraction and group activity recognition. (Therefore I have also removed Docker Compose.)
Currently,
nvidia/cuda:10.0-cudnn7-develis used as base image. I also trieddocker pull nvidia/cuda:11.0.3-cudnn8-devel(what can be built as well), but that will run into a cudaSuccess (2 vs. 0) out of memory error - according to https://github.com/CMU-Perceptual-Computing-Lab/openpose/issues/1230#issuecomment-774706454 newer cuDNN versions seem to need more memory for the same model.The container inside which everything works comes with the following details: