Yolov5 last version - architecture explanation

[Guemann-ui]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

Hi! I'm trying to understand how the last Yolo version works layer by layer from backbone to detect, currently, I'm using Yolov5n as my model, unfortunately, until now I didn't find any paper to help me to understand more the architecture, all the guides talked only about the function of each part but they didn't explain how the architecture work layer by layer and the function of each layer! Does any help please to understand all the architectural details?
Thanks.

Additional

No response

[glenn-jocher]

@besmaGuesmi yes I'm sorry we don't have a publication out yet, but the best place to start learning is to read the first 3 YOLO papers for v1-v3 by Redmon here:
https://pjreddie.com/publications/

[Guemann-ui]

Hi @glenn-jocher thank you for your reply! where can I find the internal architecture of C3/SPPF of Yolov5n? I really need to understand it! Thanks

[glenn-jocher]

@besmaGuesmi SPPF was introduced in v6.0 and is simply a faster (i.e. better implemented, eliminates redundant ops) version of SPP, the outputs are mathematically identical. See PR #4420

C3 was introduced in v4.0, it is an CSPBottleneck() layer with one of the 4 main convolutions removed, i.e. hence the name '3', bringing additional architecture simplification. See https://github.com/ultralytics/yolov5/tree/v4.0

[Guemann-ui]

Thank you!

[Guemann-ui]

Hi, @glenn-jocher I designed this architecture as SPPF according to the link that you sent me, I would check with you if there are any missing details. Thanks

[glenn-jocher]

@besmaGuesmi yes that seems correct. You can also visualize individual modules with TensorBoard once you start training. This is SPPF:

[glenn-jocher]

@besmaGuesmi a second option for module visualization is ONNX. You can export a module and visualize it with Netron:

import torch
from models.common import SPPF

torch.onnx.export(model=SPPF(128,128,5), args=torch.zeros(1,128,80,80), f='SPPF.onnx')

[Guemann-ui]

Great! Thank you @glenn-jocher

[github-actions]

👋 Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 🚀 resources:

• Wiki – https://github.com/ultralytics/yolov5/wiki
• Tutorials – https://docs.ultralytics.com/yolov5
• Docs – https://docs.ultralytics.com

Access additional Ultralytics ⚡ resources:

• Ultralytics HUB – https://ultralytics.com/hub
• Vision API – https://ultralytics.com/yolov5
• About Us – https://ultralytics.com/about
• Join Our Team – https://ultralytics.com/work
• Contact Us – https://ultralytics.com/contact

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐!