Semantic segmentation : Switch back from conda to venv , pip and dependencies ?

[jeremyfix]

• [ ] Some required libs not yet in the conda env.yml : onnxruntime-gpu , onnx
• [ ] decide on wheter to stay on conda, switch back to pip , or consider other solutions for handling dependencies (in particulier, if we which to automate testing the code base with CI/CD)

For the semantic segmenation lab, the dependencies are installed with conda, which enforces users to install conda;

We can manage all dependencies with pip and base virtualenv with requirements.txt rather than requiring someone to install the whole conda.

Note however that conda actually provides the ability to install conda-toolkit which would allow a more contenerized delivrable than using pip.