Source code for imgutils.detect.face

"""
Overview:
    Detect human faces in anime images.

    Trained on dataset `Anime Face CreateML <https://universe.roboflow.com/my-workspace-mph8o/anime-face-createml>`_
    with YOLOv8.

    .. image:: face_detect_demo.plot.py.svg
        :align: center

    This is an overall benchmark of all the face detect models:

    .. image:: face_detect_benchmark.plot.py.svg
        :align: center

    The models are hosted on
    `huggingface - deepghs/anime_face_detection <https://huggingface.co/deepghs/anime_face_detection>`_.

"""
from functools import lru_cache
from typing import List, Tuple

from huggingface_hub import hf_hub_download

from ._yolo import _image_preprocess, _data_postprocess
from ..data import ImageTyping, load_image, rgb_encode
from ..utils import open_onnx_model


@lru_cache()
def _open_face_detect_model(level: str = 's', version: str = 'v1.4'):
    return open_onnx_model(hf_hub_download(
        f'deepghs/anime_face_detection',
        f'face_detect_{version}_{level}/model.onnx'
    ))


[docs]def detect_faces(image: ImageTyping, level: str = 's', version: str = 'v1.4', max_infer_size=640, conf_threshold: float = 0.25, iou_threshold: float = 0.7) \ -> List[Tuple[Tuple[int, int, int, int], str, float]]: """ Overview: Detect human faces in anime images. :param image: Image to detect. :param level: The model level being used can be either `s` or `n`. The `n` model runs faster with smaller system overface, while the `s` model achieves higher accuracy. The default value is `s`. :param version: Version of model, default is ``v1.4``. Available versions are ``v0``, ``v1``, ``v1.3`` and ``v1.4``. :param max_infer_size: The maximum image size used for model inference, if the image size exceeds this limit, the image will be resized and used for inference. The default value is `640` pixels. :param conf_threshold: The confidence threshold, only detection results with confidence scores above this threshold will be returned. The default value is `0.25`. :param iou_threshold: The detection area coverage overlap threshold, areas with overlaps above this threshold will be discarded. The default value is `0.7`. :return: The detection results list, each item includes the detected area `(x0, y0, x1, y1)`, the target type (always `face`) and the target confidence score. Examples:: >>> from imgutils.detect import detect_faces, detection_visualize >>> >>> image = 'mostima_post.jpg' >>> result = detect_faces(image) # detect it >>> result [ ((29, 441, 204, 584), 'face', 0.7874319553375244), ((346, 59, 529, 275), 'face', 0.7510495185852051), ((606, 51, 895, 336), 'face', 0.6986488103866577) ] >>> >>> # visualize it >>> from matplotlib import pyplot as plt >>> plt.imshow(detection_visualize(image, result)) >>> plt.show() """ image = load_image(image, mode='RGB') new_image, old_size, new_size = _image_preprocess(image, max_infer_size) data = rgb_encode(new_image)[None, ...] output, = _open_face_detect_model(level, version).run(['output0'], {'images': data}) return _data_postprocess(output[0], conf_threshold, iou_threshold, old_size, new_size, ['face'])