Why YOLO11n took longer for both training and inference compared to YOLOv8n？

SMDR1412 commented 3 days ago

Search before asking

[X] I have searched the Ultralytics YOLO issues and discussions and found no similar questions.

Question

I compared YOLOv8 and YOLO11 on a fire detection dataset and found that YOLO11n took longer for both training and inference compared to YOLOv8n. In theory, YOLO11 should be faster due to having fewer parameters and lower FLOPs, but this was not the case in practice. Why is this happening?

Additional

1

UltralyticsAssistant commented 3 days ago

👋 Hello @SMDR1412, thank you for your interest in Ultralytics 🚀! We understand the curiosity surrounding different model performances.

For a detailed analysis of your issue, it's helpful to provide more context. If this is a 🐛 Bug Report, please share a minimum reproducible example that clearly demonstrates the issue you're encountering.

If this involves custom training or model-specific ❓ questions, please provide detailed information such as datasets, configurations, and any training logs you have. Ensure you're following our Tips for Best Training Results.

Join our community for more interaction:

Discord 🎧 for real-time chats
Discourse for in-depth discussions
Subreddit for community insights

Upgrade

Ensure your ultralytics package and dependencies are up-to-date to eliminate version-related issues:

pip install -U ultralytics

Environments

Verify your setup in one of the following environments, each preloaded with necessary dependencies:

Notebooks with free GPU:
Google Cloud and Amazon setups or a Docker Image, all detailed in our docs for smooth deployment.

Status

Check the CI status to ensure core functionalities operate correctly:

An Ultralytics engineer will review and assist you soon. Thank you for your patience! 😊

ambitious-octopus commented 2 days ago

Hey @SMDR1412 can I have your yolo check output, please?

Laughing-q commented 2 days ago

@SMDR1412 The speed really depends on the architecture design and the device, not just the params and flops. :) And yes yolo11n is slightly slower than yolov8 in our tests as well, however it comes with 2.1 mAP improvement and has better speed accuracy trade-off so we stick with the design. :)

ultralytics / ultralytics