Focus

[ycdhqzhiai]

❔Question

I'd like to know how much effect the removal of Focus structure will have on map. Has anyone tried it

Additional context

[github-actions[bot]]

Hello @ycdhqzhiai, thank you for your interest in our work! Please visit our Custom Training Tutorial to get started, and see our Jupyter Notebook , Docker Image, and Google Cloud Quickstart Guide for example environments.

If this is a bug report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom model or data training question, please note Ultralytics does not provide free personal support. As a leader in vision ML and AI, we do offer professional consulting, from simple expert advice up to delivery of fully customized, end-to-end production solutions for our clients, such as:

• Cloud-based AI systems operating on hundreds of HD video streams in realtime.
• Edge AI integrated into custom iOS and Android apps for realtime 30 FPS video inference.
• Custom data training, hyperparameter evolution, and model exportation to any destination.

For more information please visit https://www.ultralytics.com.

[Aktcob]

+1

[glenn-jocher]

Focus() module is designed for FLOPS reduction and speed increase, not mAP increase.

[glenn-jocher]

Also designed for layer count reduction. 1 Focus module replaces 3 yolov3/4 layers.

[captainst]

I have a question concerning the focus layer: while it reduces the computation cost, does the "stride" sampling mess the coordinates of the ground truth bounding box ?

[glenn-jocher]

@captainst Focus (and it's new v6.0 replacement) stack spatial information into channel space, so it's possible 1 pixel's worth of regression information may be reduced, though for most use cases regression accuracies will never approach 1 pixel anyway.

[captainst]

@glenn-jocher Many thanks. Now I appreciate why the focus layer is appied only once on the raw input image in the very beginning of the pipeline.

[glenn-jocher]

@captainst yes, for classification models it might be more usable later on in the backbone, but probably not for detection models without losing mAP@0.5:0.95 (mAP@0.5 would be less affected).