Ultralytics platform

Support Ultraltyics Platform
1

Hi,

I am using ultralytics platform to train my monkey database class, using the YOLO8N–CLS, all going well, i can test the ultralytics predict to certify the correct train and after converted to my SG2002 hardware, all inferences goes well, near to 100%

When i used the YOLO11N-CLS with my same monkey database, i see something weird in my hardware inference, all with values to much far to 99%, then i back to to ultralytics platform and see the same on predict :frowning:

not detects “BARBIE” monkey

and backing to the yolov8n-cls, all is good, its “BARBIE” monkey

Can you see that ?

Thanks

2

How large is your dataset? How many images do you have?

3

2Gbytes
2200 images, 44 classes

I am trainning now on COLAB to see if is a ultralytics platform bug.

PS: in some moment ultralytics platform reported “low memory” and aborted

4

Can you provide the link to your project on Ultralytics Platform?

5

Wizardly Swan Models by Miguel Wisintainer

6

Thanks for sharing the Ultralytics Platform project link — that helps.

The earlier low memory abort is the biggest clue here. That usually points more to a training/resource issue than a predict bug, similar to the Platform troubleshooting guidance. With 2200 images across 44 classes, Ultralytics YOLO11-CLS can also be a bit more sensitive than your YOLOv8n-cls run.

If your Colab test with the latest ultralytics shows the same result, then it’s probably not Platform-specific. I’d mainly verify that you exported/used best.pt and not last.pt, then compare the YOLO11 run’s top1 and confusion matrix against the YOLOv8 run. If the Platform run fails again, try the same setup with a smaller batch size.

7

Thank you!!!