FaceFinder.py 文件源码

def localize(img,model_path):
    sess = tfac.start_sess()
    y,x_hold,y_hold,keep_prob = build_net(sess)
    saver = tf.train.Saver()
    saver.restore(sess,model_path)

    #Run all detection at a fixed size
    img = cv2.resize(img,DET_SIZE)
    mask = np.zeros(img.shape)
    #Run sliding windows of different sizes
    for bx in range(WIN_MIN,WIN_MAX,WIN_STRIDE):
        by = bx
        for i in xrange(0, img.shape[1]-bx, X_STEP):
            for j in xrange(0, img.shape[0]-by, Y_STEP):
                sub_img = cv2.resize(img[i:i+bx,j:j+by],face_ds.IN_SIZE)
                X = sub_img.reshape((1,tfac.dim_prod(face_ds.IN_SIZE)))
                out = y.eval(session=sess,feed_dict={x_hold:X,keep_prob:1})[0]
                if out[0] >= CONF_THRESH:
                    mask[i:i+bx,j:j+by] = mask[i:i+bx,j:j+by]+1

    sess.close()
    mask = np.uint8(255*mask/np.max(mask))
    faces = img*(cv2.threshold(cv2.blur(mask,BLUR_DIM),0,255,cv2.THRESH_OTSU)[1]/255)
    return (faces,mask)