Table 11 Computational efficiency vs. accuracy: pruned model vs. YOLOv11s (DDTE: −9.2%, +1.9% \(\hbox {mAP}_{50}\); GC10-DET: −25.4%, +11.1% \(\hbox {mAP}_{50}\)).
Datasets | Methods | params | Gflops | \(\hbox {mAP}_{50}\) | \(\hbox {mAP}_{50-95}\) |
|---|---|---|---|---|---|
DDTE | YOLOv11s | 10.73 | 213.06 | 80.4 | 58.2 |
Prune(Ours) | 8.46 (−21.2%) | 193.51(−9.2%) | 81.9 (+1.9%) | 60.7 (+4.3%) | |
NEU-DET | YOLOv11s | 10.73 | 213.06 | 75 | 43.1 |
prune(Ours) | 8.46(−21.2%) | 158.5(−25.6%) | 77.6(+3.5%) | 45.0 (+4.4%) | |
GC10-DET | YOLOv11s | 10.73 | 213.06 | 60.4 | 29.7 |
prune(Ours) | 8.46(−21.2%) | 158.94(−25.4%) | 67.1(+11.1%) | 33.1(+11.4%) | |
PASCAL VOC 2007 | YOLOv11s | 10.73 | 213.06 | 75.2 | 55 |
prune(Ours) | 8.46(−21.2%) | 198.46(−6.9%) | 75.4(+0.3%) | 54.7(−0.5%) | |
BCCD | YOLOv11s | 10.73 | 213 | 90.8 | 62 |
prune(Ours) | 8.46(−21.2%) | 158.94(−25.4%) | 93.3(+2.8%) | 63.6(+2.6%) |
