Table 3 Ablation study results of the YOLO-SAM.
From: YOLO-SAM an end-to-end framework for efficient real time object detection and segmentation
Model Configuration | Params (M) | FLOPs (G) | mAP@0.5 | mAP@0.5:0.95 | IoU(%) | mIoU(%) | FPS |
|---|---|---|---|---|---|---|---|
YOLO-World-S | 77 | 297 | 57 | 48.3 | - | - | 373 |
+EfficientSAM | 90 | 318 | 57.3 | 48.5 | 22.9 | 75.6 | 319 |
+EfficientSAM + PSBRA | 91 | 318 | 58.1 | 48.9 | 24 | 76.6 | 315 |
+EfficientSAM + YS-CSPLayer | 91 | 318 | 57.7 | 48.6 | 23.4 | 76.1 | 315 |
+EfficientSAM + LSKA | 90 | 318 | 57.6 | 48.6 | 23.1 | 75.6 | 318 |
+EfficientSAM + PSBRA + LSKA | 92 | 318 | 58.3 | 49 | 26.5 | 82.8 | 312 |
+EfficientSAM + PSBRA + YS-CSPLayer | 93 | 318 | 58.5 | 49 | 26.2 | 82.6 | 310 |
+EfficientSAM + LSKA + YS-CSPLayer | 91 | 318 | 58 | 48.7 | 24.3 | 78.1 | 315 |
YOLO-SAM(Full Model) | 93 | 318 | 58.8 | 49.1 | 26.2 | 81.3 | 308 |
