LinasKo

Let us know if it works - that'd give more reason to revisit the PR 🙂

Could be an ultralytics version issue. Unless you need yolo11, I'd try `ultralytics==8.2.103`.

Hi @tibeoh, It's very refreshing to see a thorough test suite together with the changes. Thank you for that! I'd like to ask for a rollback of the deprecated parameter...

Also, would you mind creating a usage example in a Colab? You may use our [Starter Template](https://colab.research.google.com/drive/1rin7WrS-UvVIe-_Gfxmu-yVslGphOq89#scrollTo=pjmCrNre2g58). This eases our review considerably and creates a way for us to look...

This time - sure, let's skip the Colab. The change is small, and I can do a few tests on my end instead.

If there's contradictions in the docs, feel free to address that! We'll accept all the help we can get. Alternatively, if you lack time, let me know where the issues...

Hi @tibeoh 👋 We'd love to have this fix in the new release - there are a few outstanding changes, but I'll take over and get it merged. Thank you...

Hi @patel-zeel :wave: Thanks for giving a very specific example. That's invaluable to us. Generally - sounds good! What we need to be careful about is silent failures. For example,...

`cv2` loads the image as a `numpy` array, which makes all subsequent computation easier. I'd strongly prefer to stick with `cv2` and avoid conversions from PIL every time. Supervision uses...

Interesting details about the shape check. Thanks for carrying out the speed analysis! I'm happy for us to change the shape check to use PIL in yolo dataset case, as...