Initial commit of people detection experiment

[marcbelmont]

@danleyb2 what changes did you make to the source project, https://github.com/deepakcrk/yolov5-crowdhuman ?

[kal-tek]

failed to solve with frontend dockerfile.v0: failed to create LLB definition: failed to authorize: rpc error: code = Unknown desc = failed to fetch anonymous token: Get "https://authn.nvidia.com/token?scope=repository%3Anvidia%2Fpytorch%3Apull%2Cpush&service=registry": dial tcp: lookup authn.nvidia.com: no such host This Error happend on docker build @danleyb2. How should this be resolved?

[danleyb2]

@marcbelmont changes:

1. Stripping out code and features not needed to achieve the bare minimum of detecting people and heads in an image.
2. Newly added code is in main.py and webhook.py

Loading crowdhuman_yolov5m.pt model without it's modification requires the folder structure of models and utils to be maintained. Otherwise you'd get module import errors.

@kal-tek Try pulling the base image directly nvcr.io/nvidia/pytorch:20.12-py3 but you probably also need to setup docker authentication to nvcr.io

[marcbelmont]

@danleyb2 is there a smaller base image you could use?

[marcbelmont]

@kal-tek, is the readme enough, or should we add more details?

[kal-tek]

@kal-tek, is the readme enough, or should we add more details?

Yes the readme is enough but not everyone has nvcr docker setup. the man page seems clear but i think it can elaborate more.

also the model crowdhuman_yolov5m.pt should be added on gitignore so that it wont be registered as new file.

[danleyb2]

@marcbelmont @kal-tek Have below changes so i think the branch is good:

• Added a way to build image with CPU only support. It's smaller than with GPU. I think choosing a different base might break GPU support but i can continue looking into it.
• Added .gitignore for model file
• Updated the README to be more useful and added new info