Open
Description
Thank you for your open-source contributions! Hope your paper gets accepted soon!
I encountered an issue when using the yolov11x-obb model for large object detection: the dfl_loss fails to converge to the desired range. Visually, the detection boxes cannot fully enclose the targets. After observing the heatmaps from the backbone network, I found that the receptive field coverage in the output feature maps is insufficient to cover both ends of large objects.
I'm now looking to test YOLOv13 to address this issue, but I'm unsure:
1. Will YOLOv13x exhibit similar issues when detecting large objects?
2. If I want to adapt the OBB detection head from YOLOv11x-obb to YOLOv13x, what key considerations should I keep in mind?
Metadata
Metadata
Assignees
Labels
No labels