app.py

import cv2
import gradio as gr
import numpy as np
from mtcnn_cv2 import MTCNN

detector = MTCNN()


def predict(img, selection):
    faces = detector.detect_faces(img)

    if selection == "Low":
        opts = (anonymize_face_pixelate, 20)
    elif selection == "Medium":
        opts = (anonymize_face_pixelate, 10)
    elif selection == "High":
        opts = (anonymize_face_pixelate, 4)
    elif selection == "Emoji":
        opts = (anonymize_face_emoji, "smiley")
    else:
        raise Exception("I don't know how you did it but you chose something else.")

    if len(faces) > 0:
        for features in faces:
            img = opts[0](img, features, opts[1])
    else:
        raise Exception("No faces detected")
    return img


def anonymize_face_pixelate(image, features, blocks=10):
    bb = features["box"]
    face_crop = image[bb[1] : bb[1] + bb[3], bb[0] : bb[0] + bb[2]]
    # Divide the input image into NxN blocks
    (h, w) = face_crop.shape[:2]
    xSteps = np.linspace(0, w, blocks + 1, dtype="int")
    ySteps = np.linspace(0, h, blocks + 1, dtype="int")

    # loop over the blocks in both x and y direction
    for i in range(1, len(ySteps)):
        for j in range(1, len(xSteps)):
            # compute starting and ending (x, y)-coordinates
            # for current block
            startX = xSteps[j - 1]
            startY = ySteps[i - 1]
            endX = xSteps[j]
            endY = ySteps[i]

            # Extract the ROI using NumPy array slicing, compute the
            # mean of the ROI, and then draw a rectangle with the
            # mean RGB values over the ROI in teh original image
            roi = face_crop[startY:endY, startX:endX]
            (B, G, R) = [int(x) for x in cv2.mean(roi)[:3]]
            cv2.rectangle(face_crop, (startX, startY), (endX, endY), (B, G, R), -1)

    image[bb[1] : bb[1] + bb[3], bb[0] : bb[0] + bb[2]] = face_crop
    return image


def anonymize_face_emoji(img, features, name="smiley"):
    bb = features["box"]
    (y, x) = (bb[1] + int(bb[3] / 2), bb[0] + int(bb[2] / 2))
    (h, w) = (bb[3], bb[2])
    # Get emoji with transparency
    mask = cv2.imread("raccoon_emoji.png", -1)

    mshape = max(h, w)
    offset = int(mshape / 2)

    return overlay_transparent(img, mask, x - offset, y - offset, (mshape, mshape))


def overlay_transparent(background_img, img_to_overlay_t, x, y, overlay_size=None):
    """
    @brief      Overlays a transparant PNG onto another image using CV2

    @param      background_img    The background image
    @param      img_to_overlay_t  The transparent image to overlay (has alpha channel)
    @param      x                 x location to place the top-left corner of our overlay
    @param      y                 y location to place the top-left corner of our overlay
    @param      overlay_size      The size to scale our overlay to (tuple), no scaling if None

    @return     Background image with overlay on top
    """

    bg_img = background_img.copy()

    if overlay_size is not None:
        img_to_overlay_t = cv2.resize(img_to_overlay_t.copy(), overlay_size)

    # Extract the alpha mask of the RGBA image, convert to RGB
    b, g, r, a = cv2.split(img_to_overlay_t)
    overlay_color = cv2.merge((b, g, r))

    # Apply some simple filtering to remove edge noise
    mask = cv2.medianBlur(a, 5)

    h, w, _ = overlay_color.shape
    roi = bg_img[y : y + h, x : x + w]

    # Black-out the area behind the logo in our original ROI
    img1_bg = cv2.bitwise_and(roi.copy(), roi.copy(), mask=cv2.bitwise_not(mask))

    # Mask out the logo from the logo image.
    img2_fg = cv2.bitwise_and(overlay_color, overlay_color, mask=mask)

    # Update the original image with our new ROI
    bg_img[y : y + h, x : x + w] = cv2.add(img1_bg, img2_fg)

    return bg_img


gr.Interface(
    fn=predict,
    inputs=[
        gr.components.Image(type="numpy"),
        gr.components.Radio(["Low", "Medium", "High", "Emoji"], value="Medium"),
    ],
    outputs=gr.components.Image(type="pil"),
    allow_flagging="never",
).launch(show_error=True, quiet=False)