Make Docker Container compatible with NVIDIA runtime

mo12896 / facial-analysis-system

This repository contains the code and documentation for my master's thesis, which focuses on predicting well-being from visual data using novel AI tools for face detection, tracking, re-identification, facial emotion recognition and head pose estimation.

MIT License

2 stars 0 forks source link

Make Docker Container compatible with NVIDIA runtime #45

Closed mo12896 closed 1 year ago

mo12896 commented 1 year ago

Try out (solved):

Get Python3.10 somehow into the build (very error prone ...)
Miniconda
Multi-Stage Builds (libpython missing)
Refactor to Python 3.8.10 Compatibility (badly supported by onnx) - 0.0.11
Run with Cuda 11.8 and ubuntu:22.04 (CUDA Version change not detected by Docker!?) - 0.0.14
Rebuild docker with ubuntu:22.04 and Cuda 11.5 (need NGC-DL-CONTAINER-LICENSE file from NVIDIA repo)

mo12896 commented 1 year ago

Currently, nvidia-smi command shows, that the Container is appropriately configured. But when actually running it, it does not recognize the GPUs. Maybe try different versions. Also rebuild with cuda-11.5 path. Problem is, that neither tensorflow, pytorch nor onnxruntime detect the GPU.