Yolov5: Will you rename YOLOv5 to Fast-YOLOv4?

[Lornatang]

The work you have done is great, but in the face of doubts from the community, I can only express my sympathy.

[Louis24]

OK

[rcg12387]

Fast-YOLOv4? Is this faster than YOLOv4?

[Louis24]

Fast-YOLOv4? Is this faster than YOLOv4?

yeah

[rcg12387]

@Louis24 Then what about this?

[rcg12387]

@Louis24 Not at all. No faster, no better than YOLOv4. Besides, YOLOv4-tiny is the fastest yet has better performance than YOLOv5s.
In my opinion, at now the unique merit of YOLOv5 is that it is implemented in pytorch. Anyway, it is the on-going repo and very good news for the pytorch community including me.

[glenn-jocher]

@rcg12387 yolov4-tiny performs at 40.2 AP50 according to your link, while yolov5s (this repo) performs at 55.8 AP50. You can see this in our readme table: https://github.com/ultralytics/yolov5#pretrained-checkpoints

[rcg12387]

@glenn-jocher Ouch, I was mislead. I thought it was AP50 in your benchmark graph. Very sorry.

[glenn-jocher]

@rcg12387 thanks, its ok! YOLO (i.e. pj reddie) typically emphasized AP50 more than the AP50:95 that the community more often uses these days. We've tried to align our training for best AP50:95 mAP to more properly measure against the other detectors out there like EfficientDet, FCOS etc. that primarily use this metric.

[Kreijstal]

"YOLOv5" should be renamed YOLOtorch or something like that

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.