Fig 1.
The overall structure of DPCNet.
Fig 2.
The structure of the DPCP block.
Fig 3.
The structure of the DSFI block.
Fig 4.
The architecture of detection head.
(a) Baseline Head, (b) DPD Head.
Table 1.
Ablation experiment results on HIT-UAV dataset.
Table 2.
Targeted ablation results of cross-branch interaction in the DPD head on HIT-UAV dataset.
Table 3.
Comparison of bounding box regression loss functions within DPCNet.
Table 4.
Comparison of detection performance on the HIT-UAV test set.
Table 5.
Comparison of detection performance on the VisDrone2019 test set.
Table 6.
Runtime comparison between DPCNet and YOLO11n.
Fig 5.
Radar chart of small-object proportions in the VisDrone2019 dataset.
Fig 6.
Radar chart of small-object proportions in the HIT-UAV dataset.
Fig 7.
Recall-rate difference curves for small objects across categories on VisDrone2019.
Fig 8.
Recall-rate difference curves for small objects across categories on HIT-UAV.
Fig 9.
Small-object detection comparison on VisDrone2019.
Fig 10.
Small-object detection comparison on HIT-UAV.
Fig 11.
Heatmaps for small-object detection on VisDrone2019.
Fig 12.
Heatmaps for small-object detection on HIT-UAV.
Fig 13.
Recall improvement for small objects under low illumination on VisDrone2019.
Fig 14.
Visual comparison of small-object detection under low illumination.
Fig 15.
Recall comparison in three challenging subsets.
Fig 16.
Representative failure cases of DPCNet.
(a) Dense daytime traffic scene. (b) Low-illumination night scene.