Unable to Convert YOLOv10 Model to CoreML

THU-MIG / yolov10

YOLOv10: Real-Time End-to-End Object Detection [NeurIPS 2024]

https://arxiv.org/abs/2405.14458

GNU Affero General Public License v3.0

10.05k stars 1k forks source link

Unable to Convert YOLOv10 Model to CoreML #186

Open sajjadsarkoobi opened 6 months ago

sajjadsarkoobi commented 6 months ago

Hello YOLOv10 Team,

I have been trying to convert a YOLOv10 model to CoreML but have encountered several difficulties. I have followed various approaches that were previously successful with YOLOv5 and YOLOv8, but none have worked for YOLOv10.

Request: Could you please provide guidance on the best way to convert a YOLOv10 model to CoreML? Is it currently possible to perform this conversion? If not, are there any plans to support CoreML export in future releases?

Thank you for your assistance.

Best regards,

Hardy-Chung commented 5 months ago

希望支持导出CoreMl

Oil3 commented 3 months ago

@sajjadsarkoobi @Hardy-Chung , hi, hi, in July I had tried and could have a compilable model, objects were accurately detected however the coordinates outputted were wrong (like allways the same). I tried again, I'm not quite sure what change was different but now it seems it's all good and bounding boxes are correctly placed. It's very fast!! The CoreML "n" version is five times faster than the PyTorch "n" version. The "X" version is more than ten times faster. Compared to V8, the "n" version is half a ms slower, but the "x" version is a couple ms faster, both in CoreML.

I uploaded v10n and v10s here https://github.com/Oil3/yolov10-coreML