Table 3.

Detection results for the ellipses extracted with categorical segmentation and the combined segmentation: Regardless of the selected backbone, detection rates of about 95% (F1 score) are achieved. For detailed information about precision and recall, see Table 8. It is noticeable that, with the combined segmentation approach, the accuracy of the binary segmentation remains unaffected, although the segmentation head and the pixel-embedding head jointly influence the weights in the backbone. The experiments were carried out on all test images and separately on the daylight (D) and night vision (N) images only.

Categorical	PQ	PQ (D)	PQ (N)	F1	F1 (D)	F1 (N)	Cat. Acc	Cat. Acc (D)	Cat. Acc (N)
ResNet34	0.7920	0.8124	0.7738	0.9550	0.9619	0.9487	0.9612	0.9664	0.9565
Incep.-RN-v2	0.7943	0.8165	0.7742	0.9541	0.9614	0.9475	0.9612	0.9664	0.9564
Combined	PQ	PQ (D)	PQ (N)	F1	F1 (D)	F1 (N)	Bin. Acc	Bin. Acc (D)	Bin. Acc (N)
ResNet34	0.7966	0.8181	0.7774	0.9513	0.9588	0.9446	0.9722	0.9761	0.9687
Incep.-RN-v2	0.7921	0.8179	0.7689	0.9481	0.9566	0.9404	0.9707	0.9752	0.9666