gender_age.py

import cv2
import shutil
import numpy as np
from dataclasses import dataclass
from tqdm import tqdm
from mivolo.predictor import Predictor
from utils import *

import warnings
warnings.filterwarnings("ignore")


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


class ValidImgDetector:

    predictor = None

    def __init__(self):
        detector_path = "./model/yolov8x_person_face.pt"
        age_gender_path = "./model/model_imdb_cross_person_4.22_99.46.pth.tar"
        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, _ = 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 has_child, has_female, has_male

    def valid_img(self, img_path):
        image = cv2.imread(img_path)
        has_child, has_female, has_male = self._detect(
            image, 0.4, 0.7, "Use persons and faces", self.predictor)
        return (not has_child) and (has_female) and (not has_male)


if __name__ == "__main__":
    detector = ValidImgDetector()
    create_dir('./output/valid')
    create_dir('./output/invalid')

    for root, _, files in os.walk('./images'):
        for file in tqdm(files):
            if file.endswith('.jpg'):
                src_path = f"./images/{file}"
                dst_path = "./output/invalid"
                if detector.valid_img(src_path):
                    dst_path = "./output/valid"

                shutil.move(src_path, dst_path)