YOLOv8 performance metrics

[LOCKminiumRSY]

so i've been trying out the YOLO algortihm and i'm a bit confused with box_loss, cls_loss, and dfl_loss. i can't find the term in the ultralytics website. i would love to know what the term means and is train/box_loss different than box_loss or is it the same?? thanks

[UltralyticsAssistant]

👋 Hello @LOCKminiumRSY, thank you for your interest in Ultralytics and for trying out YOLO 🚀! It sounds like you're diving into some of the finer points of understanding model performance metrics.

For detailed explanations and clarifications, I recommend exploring the Ultralytics Docs for comprehensive information on model performance metrics. Although specific terms like box_loss, cls_loss, and dfl_loss may not be highlighted in one place, the documentation offers insights that can help interpret these metrics.

If this discussion pertains to a misunderstanding or a potential 🐛 Bug, please provide a minimum reproducible example for us to better understand the context. This step would greatly assist us in diagnosing any issues you might be facing.

In the meantime, feel free to connect with the Ultralytics community through various platforms. For real-time conversations, join us on Discord 🎧. Browse deeper discussions on Discourse, or engage with peers on our Subreddit.

Upgrade

Ensure you are using the latest release of the ultralytics package with all requirements installed in a Python>=3.8 environment and PyTorch>=1.8. Doing so can help verify that the latest improvements and fixes are applied:

pip install -U ultralytics

Environments

Experiment with YOLO in up-to-date and verified environments, complete with pre-installed dependencies like CUDA, CUDNN, Python, and PyTorch:

• Notebooks with free GPU access:
• Google Cloud Deep Learning VM: check out our GCP Quickstart Guide
• Amazon Deep Learning AMI: see our AWS Quickstart Guide
• Docker Image: refer to our Docker Quickstart Guide

Status

If this badge is green, all Ultralytics CI tests are passing. These continuous integration tests guarantee the correct functioning of YOLO across various Modes and Tasks on macOS, Windows, and Ubuntu platforms, both daily and on every commit.

This is an automated response, but rest assured, an Ultralytics engineer will assist you soon to further address your queries. 😊

[glenn-jocher]

@LOCKminiumRSY box loss measures the accuracy of predicted bounding boxes, cls loss evaluates classification accuracy, and dfl loss relates to distribution focal loss for better localization. Train/box_loss typically refers to the same concept as box_loss. For more details, please visit the Ultralytics documentation.

[glenn-jocher]

@LOCKminiumRSY you're welcome! If you have any more questions, feel free to ask.

[AKAGUZZ]

Hi @glenn-jocher,

I hope you’re doing well. I have a couple of questions about some metrics used in YOLO.

First, I've noticed that metrics like "metrics/precision(B)" and "metrics/recall(B)" include a "(B)" label, and I’d like to understand what it specifically refers to.

Additionally, I haven’t been able to locate detailed official documentation on parameters like box_loss, cls_loss, and dfl_loss. My understanding is that "train" and "val" indicate whether these losses were observed during training or validation, but any further clarification would be greatly appreciated.

Thanks for your time and all the work you’ve put into these tools!

[glenn-jocher]

The "(B)" label typically refers to metrics calculated at a specific IoU threshold, often 0.5, known as the "best" threshold. For detailed explanations of box_loss, cls_loss, and dfl_loss, these terms relate to bounding box regression, classification, and distribution focal loss, respectively, during training or validation phases. For more insights, please refer to the Ultralytics documentation.

[AKAGUZZ]

Thank you very much :)

[glenn-jocher]

You're welcome! If you have any more questions, feel free to ask.