Table 9 SOTA model experimental results on the ExDark dataset.
From: Complex dark environment-oriented object detection method based on YOLO-AS
Method | Params(M) | Flops (G) | map@50 (%) | map@50:95 (%) | Speed (ms) |
|---|---|---|---|---|---|
YOLOv3 | 61.5 | 156 | 67.60 | 39.53 | 22.20 |
YOLOv4 | 63 | 180 | 70.11 | 39.76 | 25.00 |
YOLOv5s48 | 7.2 | 16.5 | 70.34 | 40.21 | 7.30 |
YOLOv726 | 37 | 105 | 72.61 | 40.53 | 11.10 |
YOLOv8n49 | 3.2 | 8.2 | 71.67 | 42.98 | 4.00 |
YOLOv1150 | 48 | 135 | 78.47 | 44.67 | 14.30 |
DETR51 | 41 | 86 | 77.56 | 44.23 | 35.70 |
Swin Transformer | 48 | 145 | 78.94 | 44.75 | 40.00 |
Faster R-CNN52 | 137 | 207 | 77.37 | 43.95 | 55.60 |
RetinaNet53 | 36.7 | 97 | 76.19 | 43.37 | 50.00 |
EfficientDet54 | 3.9 | 2.5 | 71.95 | 43.61 | 7.90 |
YOLO-AS | 43.0 | 105.6 | 78.39 | 44.59 | 14.20 |
