Object Detect

You can track the latest updates by visiting the project's github address：Object Detect Repository

Requirement:

python >= 3.10

update

2024.9.1: support data enhancement
2024.9.2: support visual annotation file in web demo
2024.9.3: support nms in demo
2024.9.9: fix polygon

Goal

Use huggingface to implement a variety of tasks, and you can replace the model at any time without modifying the code.

Train Step:

1. python -m venv .env
2. source .env/bin/activate
3. pip install -r requirements.txt
4. modify yaml config
5. torchrun main.py (yaml_path) or python main.py

Eval Step:

python demo/inference_demo.py

WebUI

1. python demo/web_demo.py
2. open link with your browser

Note: during training, only the model file is saved, for the image pre-processing, it is not saved, you need to manually put the pre-processing configuration file into the model file to be used

FAQ

1. open too many file

ulimit -n xxx  # increase open file

2. How to download a model to train

1. open this (https://huggingface.co/models)
2. choose and download a model
3. modify yaml

3. Multi Gpu how to train

torchrun --nproc-per-node=x main.py configs/test.yaml

see more in (https://pytorch.org/docs/stable/elastic/run.html)

4. About offline, like models--timm--resnet50.a1_in1k

1. you download model
2. put this model to your ~/.cache/huggingface/hub/

note: if you want deploy this model, maybe you need add this model in docker root/.cache

Support the Author

If you find this project helpful and would like to support the author, you can make a donation using WeChat Pay or Alipay by scanning the QR codes below.

WeChat Pay:

Alipay:

Thank you for your support!