abhiWanKenobi/yolov8n_head_detection

YOLO Head Detection

YOLO-based human head detection using the HollywoodHeads dataset and Ultralytics YOLOv8.

Results

In-distribution — HollywoodHeads test split

mAP(50)	mAP50-95	precision	recall	fitness
0.871	0.529	0.894	0.796	0.529

Cross-dataset (generalisation)

Dataset	mAP(50)	mAP50-95	precision	recall	fitness
DAD-3D	0.937	0.535	0.892	0.892	0.535
300W-LP	0.855	0.265	0.827	0.883	0.265

Github Repo

You can view the training and testing scripts from here

USage

from huggingface_hub import hf_hub_download
from ultralytics import YOLO

model_path = hf_hub_download(
    repo_id="abhiWanKenobi/yolov8n-head-detection",
    filename="yolov8n_head_detector.pt",
)
model = YOLO(model_path)
results = model("image.jpg")

Acknowledgements

Datasets

• HollywoodHeads

@inproceedings{vu15heads,
    author = {Vu, Tuan{-}Hung and Osokin, Anton and Laptev, Ivan},
    title = {Context-aware {CNNs} for person head detection},
    booktitle =  {International Conference on Computer Vision (ICCV)},
    year = {2015}}

• 300W-LP

@article{DBLP:journals/corr/ZhuLLSL15,
  author    = {Xiangyu Zhu and
               Zhen Lei and
               Xiaoming Liu and
               Hailin Shi and
               Stan Z. Li},
  title     = {Face Alignment Across Large Poses: {A} 3D Solution},
  journal   = {CoRR},
  volume    = {abs/1511.07212},
  year      = {2015},
  url       = {http://arxiv.org/abs/1511.07212},
  archivePrefix = {arXiv},
  eprint    = {1511.07212},
  timestamp = {Mon, 13 Aug 2018 16:48:23 +0200},
  biburl    = {https://dblp.org/rec/bib/journals/corr/ZhuLLSL15},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

• DAD-3DHeads dataset

@inproceedings{dad3dheads,
    title={DAD-3DHeads: A Large-scale Dense, Accurate and Diverse Dataset for 3D Head Alignment from a Single Image},
    author={Martyniuk, Tetiana and Kupyn, Orest and Kurlyak, Yana and Krashenyi, Igor and Matas, Ji\v{r}i and Sharmanska, Viktoriia},
    booktitle = {Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR)},
    year={2022}
}

License

CC BY-NC 4.0 — free for research, not for commercial use. Model trained on research-only datasets (HollywoodHeads, 300W-LP, DAD-3DHeads).