from functools import lru_cache from typing import List, Tuple from huggingface_hub import hf_hub_download from imgutils.data import ImageTyping, load_image, rgb_encode from onnx_ import _open_onnx_model from plot import detection_visualize from yolo_ import _image_preprocess, _data_postprocess _FACE_MODELS = [ 'face_detect_best_n.onnx', 'face_detect_best_s.onnx', ] _DEFAULT_FACE_MODEL = _FACE_MODELS[0] @lru_cache() def _open_face_detect_model(model_name): return _open_onnx_model(hf_hub_download( 'deepghs/imgutils-models', f'face_detect/{model_name}' )) _LABELS = ['head'] def detect_faces(image: ImageTyping, level: str = 's', max_infer_size=1216, conf_threshold: float = 0.25, iou_threshold: float = 0.7) \ -> List[Tuple[Tuple[int, int, int, int], str, float]]: 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).run(['output0'], {'images': data}) return _data_postprocess(output[0], conf_threshold, iou_threshold, old_size, new_size, _LABELS) def _gr_detect_faces(image: ImageTyping, model_name: str, max_infer_size=1216, conf_threshold: float = 0.25, iou_threshold: float = 0.7): ret = detect_faces(image, model_name, max_infer_size, conf_threshold, iou_threshold) return detection_visualize(image, ret, _LABELS)