skyPeak-11n

YOLO11n fine-tuned on VisDrone2019-DET to detect people and vehicles in drone footage. Trained at 960 px so the tiny, far-away objects typical of aerial images stay visible.

Results (VisDrone val, 548 images)

Precision	Recall	mAP50	mAP50-95
0.538	0.411	0.425	0.248

Classes

pedestrian, people, bicycle, car, van, truck, tricycle, awning-tricycle, bus, motor

Usage

from huggingface_hub import hf_hub_download
from ultralytics import YOLO

model = YOLO(hf_hub_download("Polymanius/skyPeak-11n", "skyPeak-11n.pt"))
results = model.predict("drone_image.jpg", imgsz=960)  # use 960, it was trained at that size
results[0].show()

Training

• Base: yolo11n.pt (COCO-pretrained)
• Data: full VisDrone2019-DET train split (~6.5k images)
• 39 epochs, 960 px, batch 32, AMP, cosine LR, mosaic off for the last 5 epochs
• ~24 min of training on a single A100, Ultralytics 8.4.211

License

The weights are derived from Ultralytics YOLO11 (AGPL-3.0) and were trained on VisDrone, which is licensed CC BY-NC-SA 3.0 (non-commercial). Not intended for commercial use.

Citation (dataset)

Zhu et al., "Detection and Tracking Meet Drones Challenge", IEEE TPAMI, 2021.