Replies: 6 comments 2 replies
-
Hey, maybe this article has the answers to my questions: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8610656/ Please let me know if it's all correct. Thanks. |
Beta Was this translation helpful? Give feedback.
-
I agree, I'm also struggling to derivate the mathematical expression for the loss, |
Beta Was this translation helpful? Give feedback.
-
Could you please clarify which is the correct formula for Lbox @glenn-jocher as different links and different papers says different things about Yolov5 loss function? |
Beta Was this translation helpful? Give feedback.
-
@glenn-jocher ` def call(self, p, targets): # predictions, targets
I trained my custom model with your friendly tutorials, yolov5 v6.0, I want to write a report about it and explain the exact loss function but I am really confused. I'll appreciate it if you could guide |
Beta Was this translation helpful? Give feedback.
-
https://docs.ultralytics.com/yolov5/tutorials/architecture_description/#4-additional-features |
Beta Was this translation helpful? Give feedback.
-
I think this paper mentioned three losses properly that used in YOLOv8. Here is the paper link: Real-Time Flying Object Detection with YOLOv8 |
Beta Was this translation helpful? Give feedback.
-
Hi everybody, I have some doubt about the losses used in this version. I've already tried to understand something from the loss.py script, but I wasn't able.
First of all, I've found this article: https://www.nature.com/articles/s41598-021-01084-x
Where the author describes the loss in this way:
Now, looking in the issues, I noticed that the classification loss term is correct, except for the second sum, which is on the classes.
Also, the term called Lciou should be correct, and corresponds to the objectness loss.
My doubts are about the other term, here called confidence loss, that in v5 is called box loss. Looking at the loss.py file, I don't think that these terms matches. Also, I've read in the issues that this term is slightly different from the relative term in the previous yolo versions:
So I would be grateful for clarification regarding this term. Thanks in advice.
Beta Was this translation helpful? Give feedback.
All reactions