ultralytics platform yolov8-cls/yolov11-cls

[tcpipchip]

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

not detects “BARBIE” monkey

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

Can you see that ?

Thanks

PS: sorry if duplicated question

[UltralyticsAssistant]

👋 Hello @tcpipchip, thank you for reporting this and for using the Ultralytics Platform 🚀 This is an automated response to help gather the right details, and an Ultralytics engineer will also assist you soon.

We recommend a quick visit to the Docs for new users, where you can find many Python and CLI usage examples and where many common questions may already be answered.

Since this appears to be a possible 🐛 model behavior issue between YOLO8-CLS and YOLO11-CLS on the same dataset, please provide a minimum reproducible example if possible. For this type of report, it would be especially helpful to include:
-Your Ultralytics Platform project or training settings used for both models
-The exact model variants tested
-Classification metrics, confusion matrix, and training logs for both runs
-A few example images for the affected class, including “BARBIE” monkey
-The exported model format used on your SG2002 hardware
-Any preprocessing or postprocessing differences between Platform prediction and hardware inference

If this is a custom training ❓ question, please also provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

Join the Ultralytics community where it suits you best. For real-time chat, head to Discord 🎧 Prefer in-depth discussions? Check out Discourse. Or dive into threads on our Subreddit to share knowledge with the community.

Upgrade

Please upgrade to the latest ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8 to verify this is not already resolved in the latest version:

pip install -U ultralytics

Environments

YOLO may be run in any of the following up-to-date verified environments, with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled:

-Notebooks with free GPU:
-Google Cloud Deep Learning VM. See GCP Quickstart Guide
-Amazon Deep Learning AMI. See AWS Quickstart Guide
-Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLO Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

Thanks for including the screenshots 🙏 They’re helpful for showing the prediction difference between the two classification models, and with the additional details above, the team should be able to investigate more effectively.

[tcpipchip]

i can send you particulay thet monkey class database.

[glenn-jocher]

Thanks — we can’t review datasets privately, but if you share a small public sample and the exact train/export settings, plus the per-class metrics or confusion matrix and text prediction output for the same image, we can check it here; a minimum reproducible example is the fastest path, and for a quick sanity check I’d also try the same data with Ultralytics YOLO26 classification.

[tcpipchip]

https://drive.google.com/file/d/1JgfaV5m7JGyrfLMR0bUpZynCBvV81YYY/view?usp=sharing
44 class

[tcpipchip]

success trainning on COLAB yolo11n-cls

[glenn-jocher]

Thanks for confirming yolo11n-cls works in Colab; that suggests an Ultralytics Platform/export-side issue or a class-order mismatch, so please verify the exported model keeps the same model.names order and open a new report in GitHub Issues with the Platform run settings and a small sample per our MRE guide so we can reproduce it.

[tcpipchip]

please, test with that google drive...and using the yolo11n default values on ultralytics platform

[glenn-jocher]

Thanks for reporting it — I can’t test from a Google Drive link here, but if YOLO11n-cls is consistently missing BARBIE in Ultralytics Platform Predict, please attach a small public sample for that class plus the exact training settings and validation/confusion-matrix results so the team can reproduce it; lower confidence can vary across models, but a repeatable wrong top1 prediction is worth investigating, and our Minimum Reproducible Example guide shows the best format.

[Y-T-G]

What's the accuracy you achieved with YOLO11 training? It seems like overfitting which can occur if your training dataset is small.

[tcpipchip]

What's the accuracy you achieved with YOLO11 training? It seems like overfitting which can occur if your training dataset is small.

low, but get bigger on colab

[tcpipchip]

wow, i dont have more this info! Sorry!

[Y-T-G]

Can you post the link to your Platform projects for both YOLO11 and YOLO26?