YOLOv11n model is much less accurate in Hub app than in dashboard preview

[oliverlibaw]

Search before asking

• [X] I have searched the HUB issues and discussions and found no similar questions.

Question

My custom cat-detection model has good performance metrics in the Hub dashboard and performs quite well when I use the Preview functionality. But in the iphone app, it's extremely inaccurate (all sorts of false positives and false negatives).

https://hub.ultralytics.com/models/qZhsQXwHwz8QUTyrPc1p

Maybe I'm missing something obvious here? I see the metrics in the app don't match those in the Hub dashboard.

Additional

No response

[UltralyticsAssistant]

👋 Hello @oliverlibaw, thank you for raising an issue about the Ultralytics HUB 🚀! An Ultralytics engineer will review your issue and provide further assistance soon.

In the meantime, here are some resources that might help you troubleshoot the issue with your YOLOv11n model:

• Models: Training and Exporting. This section provides guidance on how to ensure your model is properly configured for deployment across different platforms.
• Ultralytics HUB App. Learn more about model deployment on iOS and Android, which might offer insights into the performance differences you're experiencing.
  • iOS. Understand the specifics of running YOLO CoreML models on iPhones, which could relate to your observed discrepancies.

If this is a 🐛 Bug Report, please ensure you provide a minimum reproducible example (MRE) as per our MRE guide. Any additional screenshots showing the comparison between the Hub dashboard and the app will also be helpful.

If this is a ❓ Question, additional context regarding your dataset, model parameters, and environment details will assist us in providing a more targeted response.

We appreciate your patience while we investigate this! 🚀🔍