Downloads · 30 days
0
uwaisasghar/aerial-vehicle-detection-rfdetr
aerial-vehicle-detection-rfdetr is a object detection model from uwaisasghar. Use it when you need objects located in an image. It is set up for rfdetr.
Fine-tuned RF-DETR Small checkpoint for detecting vehicles in drone and elevated-camera imagery. The model predicts one broad vehicle class covering cars, vans, trucks, buses/freight vehicles, and motorcycles.
Downloads · 30 days
0
Access
Public
Updated Sep 3, 2026
Repo size
129 MB
Likes
4
Public
Click a slice to open those files.
.pth129 MB · 100%
From the Hugging Face model README
Fine-tuned RF-DETR Small checkpoint for detecting vehicles in drone and
elevated-camera imagery. The model predicts one broad vehicle class covering
cars, vans, trucks, buses/freight vehicles, and motorcycles.
Evaluated once on a held-out, sequence-aware test split containing 3,083 images and 52,561 vehicle boxes:
| Metric | Score |
|---|---|
| COCO mAP@50:95 | 0.6477 |
| mAP@50 | 0.9180 |
| mAP@75 | 0.7498 |
| mAR@500 | 0.7082 |
| F1 (threshold sweep) | 0.8905 |
| Precision | 0.9247 |
| Recall | 0.8587 |
hf download uwaisasghar/aerial-vehicle-detection-rfdetr \
checkpoint_best_total.pth \
--local-dir runs/rfdetr_small_aerial_vehicle
SHA-256:
b7d80101d36bce349a09e3ee2965bc99195eafcb73054dd552e4b486d909202e
Clone the source repository, install its requirements, download the checkpoint as shown above, and run:
python infer_rfdetr_video.py \
--source input.mp4 \
--output output_tracked.mp4 \
--weights runs/rfdetr_small_aerial_vehicle/checkpoint_best_total.pth \
--mode hybrid \
--track-threshold 0.50 \
--box-style corner
The unified one-class dataset uses VisDrone2019-DET train/validation, a YOLO-format UAVDT repack, and DroneVehicle RGB train/validation. Capture or video sequences are kept in a single split to reduce leakage. See the source repository README for dataset links, exact preparation steps, and directory layout. Upstream datasets are not redistributed; users must review and comply with their respective terms.