search

GeekAlexis / FastMOT

High-performance multiple object tracking based on YOLO, Deep SORT, and KLT 🚀
MIT License
1.12k stars 256 forks source link

YOLOv8 support #267

Open jpfleischer opened 5 months ago

jpfleischer commented 5 months ago

How to integrate YOLOv8?

I converted onnx to trt but the fastmot asks for anchor values. YOLOv8 is anchorless.

AbdulmueezEmiola commented 2 months ago

Any update on this?

GeekAlexis commented 2 months ago

You can create a new YOLOAnchorless base class instead of subclassing the existing one: https://github.com/GeekAlexis/FastMOT/blob/32c217a7d289f15a3bb0c1820982df947c82a650/fastmot/models/yolo.py#L11

Make necessary modifications as needed. Pull requests or contributions are welcome

jpfleischer commented 2 months ago

@GeekAlexis please share a link to a discord or a platform where we can collaborate. it would be much appreciated if we could work together to improve FastMOT.

GeekAlexis commented 2 months ago

@GeekAlexis please share a link to a discord or a platform where we can collaborate. it would be much appreciated if we could work together to improve FastMOT.

Created a discord server: https://discord.gg/7CB6q9j4. Feel free to join.

chulyoung-code commented 6 days ago

@GeekAlexis @jpfleischer

Hello

I am also working on changing fastmot's detection model to yolov8. At first, I applied the yolov8 model using pytorch without trt conversion, and the detection accuracy was good, but the problem was that it was slow because it was not trt conversion.

So I'm looking for a way to convert the yolov8 model to a trt model within the fastmot pipeline.

Have you found a way? I want to share information with you I also want to help in whatever way I can.