Problem in hand detection module...

[pratik2394]

When I try to run hand detection module or hand+ body detection module, it gives the following error message...

(Body module alone works fine.)

Command: xvfb-run -a python -m demo.demo_frankmocap --input_path ./sample_data/han_short.mp4 --out_dir ./mocap_output

Loading Body Pose Estimator Loading Third View Hand Detector Pass 'Combine redundant instructions' is not initialized. Verify if there is a pass dependency cycle. Required Passes: Segmentation fault (core dumped)

Computer background:

I have RTX 4060, Cuda 10.1, 11.3, 11.8, 12.1 (passing cuda 11.3 for this environment.) I have installed torch=1.10 and detectron2 using the following command:

python -m pip install detectron2 -f \ https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/index.html

While building using install_frankmocap.sh it gives a lot of deprecation warning and deprecated messages...

Any help is appreciated.

[penincillin]

@pratik2394 Hi, the bug seems to be caused by detectron2. I would suggest you to run some example code of detectron2 after installing it.

[pratik2394]

@penincillin Thank you for replying...

I solved the issue by changing the python version.

Below are the commands that worked for me.

Changes from the installation instructions:

conda create -n venv_frankmocap python=3.9

conda install cudatoolkit=11.3 conda install cudnn=8.2.0

In addition to the above I also installed CUDA 11.3 from the official source and provided nvcc path in the .bashrc file.

Torch version 1.10

detectron2 installation:

python -m pip install detectron2 -f \ https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/index.html

[penincillin]

Thanks for sharing the experience!