Figure 1.

Figure 2.

Figure 3.

Figure 4.

Ablation Study of DM-YOLOV12 Architectural Configurations on TWDD Dataset_
| Model Configuration | YOLOv12 Scale | DINOV Variant | mAP50 |
|---|---|---|---|
| YOLOv12(baseline) | L | ViT-B/16 | 0.4325 |
| Y OLOv 12+Mamba | L | ViT-B/16 | 0.4673 |
| YOLOv 12+Mamba+DIN OV3 | L | ViT-B/16 | 0.5497 |
| YOLOv12(baseline) | L | ViT-L/16 | 0.4461 |
| Y OLOv 12+Mamba | L | ViT-L/16 | 0.4871 |
| YOLOv 12+Mamba+DIN OV3 | L | ViT-L/16 | 0.5831 |
| YOLOv12(baseline) | M | ViT-B/16 | 0.4152 |
| Y OLOv 12+Mamba | M | ViT-B/16 | 0.4367 |
| YOLOv 12+Mamba+DIN OV3 | M | ViT-B/16 | 0.5579 |
| YOLOv12(baseline) | M | ViT-L/16 | 0.4356 |
| Y OLOv 12+Mamba | M | ViT-L/16 | 0.4651 |
| YOLOv 12+Mamba+DIN OV3 | M | ViT-L/16 | 0.5878 |
| YOLOv12(baseline) | S | ViT-B/16 | 0.4051 |
| Y OLOv 12+Mamba | S | ViT-B/16 | 0.4571 |
| YOLOv 12+Mamba+DIN OV3 | S | ViT-B/16 | 0.5481 |
| YOLOv12(baseline) | S | ViT-L/16 | 0.4517 |
| Y OLOv 12+Mamba | S | ViT-L/16 | 0.4401 |
| YOLOv 12+Mamba+DIN OV3 | S | ViT-L/16 | 0.5591 |
Quantitative comparison with state-of-the-art methods_
| Model | Param.(M) | mAP50 |
|---|---|---|
| YOLOv3-M | 32.1 | 0.3425 |
| YOLOv5-M | 36.1 | 0.3973 |
| YOLOv7-M | 34.9 | 0.4197 |
| Gold-YOLO-M | 41.3 | 0.4316 |
| YOLOv8-M | 25.9 | 0.5054 |
| YOLOv10-M | 24.8 | 0.4098 |
| YOLOv12-M | 17.1 | 0.5216 |
| MD-YOLOv12-M | 16.4 | 0.5878 |