How to do object detection using burn

[Raj2032]

I know there is yolox however since there are no native implementation for training yolox-burn I was wondering if there are other plugis etc I could use with burn to first do training and then inference?

[laggui]

There is no official example that trains an object detection model right now. You would have to write the code, unless someone else has contributed such a project somewhere online.

For YOLOX specifically, there is an outstanding issue: tracel-ai/models#63

We support loading pre-trained models, either via ONNX (codegen) or pytorch/safetensors.

[antimora]

Using ONNX files, here are couple of examples of Yolo and rf-detr:

1. YOLO: https://github.com/tracel-ai/burn-onnx/tree/main/crates/model-checks/yolo
2. rf-detr: https://github.com/tracel-ai/burn-onnx/tree/main/crates/model-checks/rf-detr

[Raj2032]

@laggui I hope sometime in the near future there would be support for training.

WHat about candle even though its not for burn?

[Raj2032]

@antimora Thanks will look into it.

[Raj2032]

Hi @antimora , thanks for providing the links I will stick with the first link.

if I ran this:

python get_model.py --model yolov8n

I am not sure does "Download and prepare a model" exactly do? I assume we are not training our own custom model?

[laggui]

I hope sometime in the near future there would be support for training.

WHat about candle even though its not for burn?

What about candle, exactly? 😅 Not sure I understand what you are trying to refer to. It is supported as a backend, though it's on the way to being deprecated since it does not support many of the standard tensor module ops. Other backends perform better and have wider support.

I am not sure does "Download and prepare a model" exactly do? I assume we are not training our own custom model?

The script automatically downloads the pretrained pytorch model from the web, and exports it as ONNX.