Hello

[glenn-jocher]

testing

[glenn-jocher]

Hello 3

[hbagherpour84]

Dear Glenn jocher
I recently train yolo 8 for my project and its results were good. but in this project I want to calculate evaluation parameters such as recall, precision and accuracy. because the confusion matrix has background too, I could not understand how to calculate these parameters. does background use in formula of precision and etc.?

[glenn-jocher]

For YOLO8 evaluation metrics with background class:

1. The background class typically isn't included when calculating precision/recall metrics for object detection, since we're primarily interested in actual object detection performance.

2. For calculating metrics:

• Precision = TP / (TP + FP) for each actual object class
• Recall = TP / (TP + FN) for each actual object class
• mAP calculation focuses on detected objects vs ground truth

Given this appears to be a technical question about YOLOv8 implementation, I'd recommend opening an issue on the Ultralytics GitHub repository at https://github.com/ultralytics/ultralytics.

[hbagherpour84]

Performance evaluation of YOLO v5 model for automatic crop and weed .pdf <https://drive.google.com/file/d/1uOvDAQ7s5bfRvuIlRyD-qC1xhxhIUoQv/view?usp=drive_web> *Subject:* Clarification on Precision and Recall Calculations Dear Glenn Jocher, Thank you very much for your reply. I greatly appreciate your time and assistance. However, I am still confused about the application of the formulas for calculating precision and recall. In the file ( https://www.sciencedirect.com/science/article/pii/S2772375523000618) and the attached accompanying images, I am unable to understand how the authors applied these formulas. Could you kindly help me comprehend how they derived these parameters from their confusion matrix? Your guidance would mean a lot to me. Best regards,

…

On Mon, Nov 18, 2024 at 8:31 PM Glenn Jocher ***@***.***> wrote: For YOLO8 evaluation metrics with background class: 1. The background class typically isn't included when calculating precision/recall metrics for object detection, since we're primarily interested in actual object detection performance. 2. For calculating metrics: - Precision = TP / (TP + FP) for each actual object class - Recall = TP / (TP + FN) for each actual object class - mAP calculation focuses on detected objects vs ground truth Given this appears to be a technical question about YOLOv8 implementation, I'd recommend opening an issue on the Ultralytics GitHub repository at https://github.com/ultralytics/ultralytics. — Reply to this email directly, view it on GitHub <#142 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BGQLCYN25EHKNP5ME5JSIA32BIMQFAVCNFSM6AAAAABRB2YEQ6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIOBTGYYDQMRZG4> . You are receiving this because you commented.Message ID: ***@***.***>

-- Hossein Bagherpour, Assistant Professor, Dept of Biosystem Engineering, Bu-Ali Sina University, Hamedan, Iran. Tel: 0098-81-34425400