app.py

import gradio as gr
import cv2
from utils.detect import create_mtcnn_net, MtcnnDetector
from utils.vision import vis_face
import argparse
from PIL import Image
import numpy as np

MIN_FACE_SIZE = 24

def parse_args():
    parser = argparse.ArgumentParser(description='Test MTCNN',
                                     formatter_class=argparse.ArgumentDefaultsHelpFormatter)

    parser.add_argument('--net', default='onet', help='which net to show', type=str)
    parser.add_argument('--pnet_path', default="./model_store/pnet_epoch_20.pt",help='path to pnet model', type=str)
    parser.add_argument('--rnet_path', default="./model_store/rnet_epoch_20.pt",help='path to rnet model', type=str)
    parser.add_argument('--onet_path', default="./model_store/onet_epoch_20.pt",help='path to onet model', type=str)
    parser.add_argument('--path', default="./img/mid.png",help='path to image', type=str)
    parser.add_argument('--min_face_size', default=MIN_FACE_SIZE,help='min face size', type=int)
    parser.add_argument('--use_cuda', default=False,help='use cuda', type=bool)
    parser.add_argument('--thresh', default='[0.6, 0.7, 0.7]',help='thresh', type=str)
    parser.add_argument('--save_name', default="result.jpg",help='save name', type=str)
    parser.add_argument('--input_mode', default=1,help='image or video', type=int)
    args = parser.parse_args()
    return args
def greet(请上传待检测人脸图片):
    args = parse_args()
    thresh = [float(i) for i in (args.thresh).split('[')[1].split(']')[0].split(',')]
    pnet, rnet, onet = create_mtcnn_net(p_model_path=args.pnet_path, r_model_path=args.rnet_path,o_model_path=args.onet_path, use_cuda=args.use_cuda)
    mtcnn_detector = MtcnnDetector(pnet=pnet, rnet=rnet, onet=onet, min_face_size=args.min_face_size,threshold=thresh)
    img = cv2.imread(请上传待检测人脸图片)
    b,g,r = cv2.split(img)
    img_bg = cv2.merge([r,g,b])
    p_bboxs, r_bboxs, bboxs, landmarks = mtcnn_detector.detect_face(img)
    save_name = args.save_name
    fig = vis_face(img_bg, bboxs, landmarks, MIN_FACE_SIZE, save_name)
    fig.canvas.draw()

    # Get the RGB buffer from the figure
    w, h = fig.canvas.get_width_height()
    buf = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8)
    buf.shape = (h, w, 3)

    # canvas.tostring_rgb give pixmap in RGB mode. 
    # Roll the ALPHA channel to have it in RGBA mode
    buf = np.roll(buf, 3, axis=2)
    return Image.fromarray(buf)
iface = gr.Interface(fn=greet, 
                    inputs=gr.Image(type="filepath"),
                    outputs="image")
iface.launch()