app.py

import cv2
import numpy as np
import gradio as gr
from dataclasses import dataclass
from mivolo.predictor import Predictor
from utils import *

import warnings
warnings.filterwarnings("ignore")


@dataclass
class Cfg:
    detector_weights: str
    checkpoint: str
    device: str = "cpu"
    with_persons: bool = True
    disable_faces: bool = False
    draw: bool = True


class ValidImgDetector:

    predictor = None

    def __init__(self):
        # Use a pipeline as a high-level helper
        detector_file = "yolov8x_person_face.pt"
        age_gender_file = "model_imdb_cross_person_4.22_99.46.pth.tar"
        detector_path = f"./model/{detector_file}"
        age_gender_path = f"./model/{age_gender_file}"
        domain = "https://www.modelscope.cn/api/v1/models/monetjoe/human-detector/repo?Revision=master&FilePath="
        download_file(f"{domain}{detector_file}", detector_path)
        download_file(f"{domain}{age_gender_file}", age_gender_path)

        predictor_cfg = Cfg(detector_path, age_gender_path)
        self.predictor = Predictor(predictor_cfg)

    def _detect(
        self,
        image: np.ndarray,
        score_threshold: float,
        iou_threshold: float,
        mode: str,
        predictor: Predictor
    ) -> np.ndarray:
        # input is rgb image, output must be rgb too
        predictor.detector.detector_kwargs['conf'] = score_threshold
        predictor.detector.detector_kwargs['iou'] = iou_threshold

        if mode == "Use persons and faces":
            use_persons = True
            disable_faces = False
        elif mode == "Use persons only":
            use_persons = True
            disable_faces = True
        elif mode == "Use faces only":
            use_persons = False
            disable_faces = False

        predictor.age_gender_model.meta.use_persons = use_persons
        predictor.age_gender_model.meta.disable_faces = disable_faces

        # image = image[:, :, ::-1]  # RGB -> BGR
        detected_objects, out_im = predictor.recognize(image)

        has_child, has_female, has_male = False, False, False
        if len(detected_objects.ages) > 0:
            has_child = min(detected_objects.ages) < 18
            has_female = 'female' in detected_objects.genders
            has_male = 'male' in detected_objects.genders

        return out_im[:, :, ::-1], has_child, has_female, has_male

    def valid_img(self, img_path):
        image = cv2.imread(img_path)
        return self._detect(
            image,
            0.4,
            0.7,
            "Use persons and faces",
            self.predictor
        )


def inference(photo):
    detector = ValidImgDetector()
    return detector.valid_img(photo)


iface = gr.Interface(
    fn=inference,
    inputs=gr.Image(label='上传照片', type='filepath'),
    outputs=[
        gr.Image(label='检测结果', type='numpy'),
        gr.Textbox(label='存在儿童'),
        gr.Textbox(label='存在女性'),
        gr.Textbox(label='存在男性')
    ],
    examples=get_jpg_files('./examples')
)

# 启动 Gradio 界面
iface.launch()