Table 4 Experimental results of different SOTA algorithms.
From: Wildlife target detection based on improved YOLOX-s network
Target detection algorithms | image size | Params/M | FLOPs/G | mAP@0.5 | mAP@0.5:0.95 |
|---|---|---|---|---|---|
YOLOX-s(base) | 640 × 640 | 9.0 | 26.8 | 79.9 | 56.7 |
+ Focal-IoU Loss (ours) | 640 × 640 | 9.0 | 26.8 | 80.5 | 56.4 |
+ Focal-IoU Loss + MobileViT-Pooling (ours) | 640 × 640 | 22.4 | 51.2 | 85.4 | 60.2 |
+ Focal-IoU Loss + MobileViT-Pooling + 2× DyHead(ours) | 640 × 640 | 23.0 | 56.4 | 87.8 | 62.0 |
YOLOX-m | 640 × 640 | 25.3 | 73.8 | 81.9 | 58.5 |
YOLOX-l | 640 × 640 | 54.2 | 155.6 | 82.4 | 59.2 |
YOLOv5-m | 640 × 640 | 21.2 | 49.0 | 84.3 | 60.3 |
RetinaNet + ResNet5028 | 1280 × 800 | 36.3 | 207.9 | 84.8 | 60.7 |
DETR31 | 1280 × 800 | 41.3 | 86.0 | 77.4 | 52.6 |
Faster RCNN + ResNet5021 | 1280 × 800 | 41.2 | 206.7 | 86.2 | 61.4 |
