Skip to main content
. 2025 Sep 12;8:1637842. doi: 10.3389/frai.2025.1637842

Table 10.

Ablation experiment results.

Model mAP@0.5 (95% CI) mAP@0.95 (95% CI) P R Wsz Latency
YOLOv10 95.61% (95.29–95.93) 74.77% (74.11–75.45) 92.17% 92.18% 5.51 MB 35.71 ms
YOLOv11 94.88% (94.53–95.22) 65.13% (64.21–66.01) 92.62% 89.57% 5.30 MB 35.66 ms
YOLOv5 91.26% (90.88–91.63) 57.69% (56.79–58.60) 88.53% 83.89% 7.25 MB 41.2 ms
YOLOv8 95.39% (95.05–95.72) 62.38% (61.44–63.32) 89.79% 90.00% 5.49 MB 37.8 ms
YOLOv9-t 95.35% (95.01–95.68) 71.27% (70.35–72.16) 97.66% 86.52% 5.98 MB 39.5 ms
SSD 93.33% (93.00–93.64) 69.20% (68.29–70.09) 90.75% 93.64% 14.30 MB 54.7 ms
Full model (all modules) 97.69% (97.46–97.91) 79.39% (78.83–79.94) 94.41% 95.65% 5.37 MB 32.02 ms

The meaning of the 95% CI is as follows: under the same data and analysis procedure, if the study were repeated and the interval recalculated each time, about 95% of such intervals would cover the true value of the fold-averaged metric. This interval reflects the uncertainty arising from sample and fold-to-fold variation; it is not equivalent to an individual prediction interval, nor does it guarantee external generalization.

Values shown in bold are from the Full model with all modules enabled.