Default examples produce useless pose estimations in Ubuntu 20.04

[ThomasA]

Issue Summary

OpenPose produces gibberish pose estimations when running the OpenPose Quick Start examples (https://cmu-perceptual-computing-lab.github.io/openpose/web/html/doc/md_doc_01_demo.html).

Executed Command (if any)

build/examples/openpose/openpose.bin --image_dir examples/media/ --write_images ~/tmp/openpose_img_out/

OpenPose Output (if any)

It produces output images similar to this, i.e. with pose estimations drawn all over the place in a way that bears no resemblance to the persons in the image:

Errors (if any)

OpenPose runs successfully without error messages.

Type of Issue

• Other (output error)

Your System Configuration

1. Whole console output

   Starting OpenPose demo...
   Configuring OpenPose...
   Starting thread(s)...
   ---------------------------------- WARNING ----------------------------------
   We have introduced an additional boost in accuracy in the CUDA version of about 0.2% with respect to the CPU/OpenCL versions. We will not port this to CPU given the considerable slow down in speed it would add to it. Nevertheless, this accuracy boost is almost insignificant so the CPU/OpenCL versions can be safely used.
   -------------------------------- END WARNING --------------------------------
   OpenPose demo successfully finished. Total time: 196.564805 seconds.
   

2. OpenPose version: GitHub commit 10f25a129985c54367704ca98e6549c3d07be5bd

3. General configuration:

   • Installation mode: CMake, sh script, manual Makefile installation, ... (Ubuntu)
   • Operating system (lsb_release -a in Ubuntu):

     Distributor ID:	Ubuntu
     Description:	Ubuntu 20.04.3 LTS
     Release:	20.04
     Codename:	focal
     

   • Operating system version Ubuntu 20.04.3 LTS
   • Release or Debug mode? (by default: release): Not attempted changed, so I assume release.
   • Compiler (gcc --version in Ubuntu): gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0

4. Non-default settings:

   • 3-D Reconstruction module added? (by default: no): Not attempted changed, so I assume no.
   • Any other custom CMake configuration with respect to the default version? (by default: no): GPU_MODE=CPU_ONLY, USE_MKL un-checked, USE_CUDNN un-checked.

5. 3rd-party software:

   • Caffe version: Default from OpenPose
   • CMake version (cmake --version in Ubuntu): 3.16.3
   • OpenCV version: pre-compiled apt-get install libopencv-dev 4.2.0+dfsg-5

6. If GPU mode issue:

   • CUDA version (cat /usr/local/cuda/version.txt in most cases): I have CUDA 11.5.1, but I have disabled it in cmake due to out-of-memory error when running on GPU.
   • cuDNN version: I have cuDNN 8.3.1.22, but I have disabled it in cmake due to error when running on GPU.
   • GPU model (nvidia-smi in Ubuntu): NVIDIA T500, but not involved in the described error.

7. If CPU-only mode issue:

   • CPU brand & model: Intel(R) Core(TM) i7-1165G7 @ 2.80GHz
   • Total RAM memory available: 16GB

[gineshidalgo99]

I'd remove and re-download the pose models, it seems like the models are corrupted.

[ThomasA]

I have now deleted all files in 'models/pose' and run git checkout models/pose to re-create the .prototxt files. After this, I have re-run "Configure" and "Generate" in the cmake GUI. This re-downloaded 'models/pose/body_25/pose_iter_584000.caffemodel'. Finally, I re-ran make in the 'build' dir. OpenPose produces the same nonsense posture estimates afterwards.

[ThomasA]

I made another attempt, this time deleting all contents of 'models'. I then checked out 'models' again and ran 'models/getModels.sh' which downloaded more models. I re-ran make in the 'build' dir and it still produces the same nonsense.