def main(photo):
# Instantiate face detection neural network and load pretrained weights
model = FCN()
model.load_weights("Keras_FCN8s_face_seg_YuvalNirkin.h5")
# ~~~Process Image~~~ #
def preprocess(image):
# Resize image to 500x500 pixels
image = cv2.resize(image, (500, 500))
# Arbitrary color scaling that makes the algorithm work better on select data
image = image.astype(float) - np.array(
(122.67891434, 116.66876762, 104.00698793))
# Convert image to convention required by face detection algorithm
image = image[np.newaxis, :, :, ::-1]
return image
# Pass image through face detection algorithm
im = cv2.cvtColor(cv2.imread(photo), cv2.COLOR_BGR2RGB)
image_in = preprocess(im)
out = model.predict(image_in)
# Generate mask for face detection
out_resized = cv2.resize(np.squeeze(out), (im.shape[1], im.shape[0]))
out_resized_clipped = np.clip(out_resized.argmax(axis=2), 0,
1).astype(np.float64)
face_mask = cv2.GaussianBlur(out_resized_clipped, (7, 7), 6)
# Redness criterion of each pixel (absolute)
red_absolute = 200 > im[:, :, 0]
# Redness criterion of each pixel (relative)
red_relative = im[:, :, 0] > (0.9 * (im[:, :, 1] + im[:, :, 2]))
# Pixels must both have a high red value and relatively lower blue and green values
red_spots = np.logical_and(red_absolute, red_relative)
# Generate mask of red regions from face mask
red_mask = np.copy(face_mask)
red_mask[np.invert(red_spots)] = 0
red_mask[red_mask != 1] = 0
# Calculate redness score
face_pixels = sum(face_mask.reshape(-1) > 0)
red_pixels = sum(red_mask.reshape(-1) > 0)
red_score = 1000 * red_pixels / face_pixels
# Generate labeled image
overlay = np.copy(im)
overlay[red_mask == 1, 0] = 0
overlay[red_mask == 1, 1] = 0
overlay[red_mask == 1, 2] = 255
im_mask = cv2.addWeighted(im, 0.5, overlay, 0.5, 0)
# Output image and redness score
cv2.imwrite(photo[:-4] + '_labeled' + photo[-4:], im_mask[:, :, ::-1])
print('Redness Score: %d' % red_score)
import gradio as gr
import os, sys
file_folder = os.path.dirname(os.path.abspath(__file__))
sys.path.insert(0, file_folder)
from FCN8s_keras import FCN
from PIL import Image
import cv2
from drive import download_file_from_google_drive
import numpy as np
weights = os.path.join(file_folder, "face_seg_model_weights.h5")
if not os.path.exists(weights):
file_id = "1IerDF2DQqmJWqyvxYZOICJT1eThnG8WR"
download_file_from_google_drive(file_id, weights)
model1 = FCN()
model1.load_weights(weights)
def segment_face(inp):
im = Image.fromarray(np.uint8(inp))
im = im.resize((500, 500))
in_ = np.array(im, dtype=np.float32)
in_ = in_[:, :, ::-1]
in_ -= np.array((104.00698793, 116.66876762, 122.67891434))
in_ = in_[np.newaxis, :]
out = model1.predict(in_)
out_resized = cv2.resize(np.squeeze(out), (inp.shape[1], inp.shape[0]))
out_resized_clipped = np.clip(out_resized.argmax(axis=2), 0,
1).astype(np.float64)
#predict facial kepoints
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')
dets = detector(img, 1)
parts = predictor(img, dets[0]).parts()
dlib_pts = []
for pt in parts:
dlib_pts.append(np.array([pt.x, pt.y]))
dlib_pts = np.array(dlib_pts)
#mask from triangulation
face_triangles = make_face_triang(dlib_pts, (H, W))
triang_mask = get_triang_mask(img.shape, face_triangles)
# NN segmentation
seg_model = FCN()
seg_model.load_weights("Keras_FCN8s_face_seg_YuvalNirkin.h5")
nn_mask = get_nn_mask(img, seg_model)
K.clear_session()
#get hairline from NN mask
contour = find_contours(nn_mask[:, :, 0], 0.5)[0].astype(np.int16)
contour = contour[np.logical_or(
np.logical_and(
contour[:, 0] < dlib_pts[0][1],
np.abs(contour[:, 1] - dlib_pts[0][0]) <=
np.abs(contour[:, 1] - dlib_pts[16][0])),
np.logical_and(
contour[:, 0] < dlib_pts[16][1],
np.abs(contour[:, 1] - dlib_pts[0][0]) >
np.abs(contour[:, 1] - dlib_pts[16][0])))]
from keras.layers import *
from keras.layers.advanced_activations import LeakyReLU
from keras.activations import relu
from keras.initializers import RandomNormal
from keras.applications import *
import keras.backend as K
import numpy as np
import os
import matplotlib.pyplot as plt
import cv2
import scipy.ndimage as ndimage
import time
from FCN8s_keras import FCN
model = FCN()
model.load_weights("Keras_FCN8s_face_seg_YuvalNirkin.h5")
def vgg_preprocess(im):
im = cv2.resize(im, (500, 500))
in_ = np.array(im, dtype=np.float32)
in_ = in_[:, :, ::-1]
in_ -= np.array((104.00698793, 116.66876762, 122.67891434))
in_ = in_[np.newaxis, :]
return in_
def auto_downscaling(im):
w = im.shape[1]
h = im.shape[0]
import gradio as gr
import os, sys
sys.path.insert(0, os.path.join(os.path.dirname(os.path.abspath(__file__), "utils"))
from FCN8s_keras import FCN
from PIL import Image
import cv2
import tensorflow as tf
from drive import download_file_from_google_drive
import numpy as np
weights = os.path.join(file_folder, "face_seg_model_weights.h5")
if not os.path.exists(weights):
file_id = "1IerDF2DQqmJWqyvxYZOICJT1eThnG8WR"
download_file_from_google_drive(file_id, weights)
model1 = FCN()
model1.load_weights(weights)
def segment_face(inp):
im = Image.fromarray(np.uint8(inp))
im = im.resize((500, 500))
in_ = np.array(im, dtype=np.float32)
in_ = in_[:, :, ::-1]
in_ -= np.array((104.00698793,116.66876762,122.67891434))
in_ = in_[np.newaxis,:]
out = model1.predict(in_)
out_resized = cv2.resize(np.squeeze(out), (inp.shape[1], inp.shape[0]))
out_resized_clipped = np.clip(out_resized.argmax(axis=2), 0, 1).astype(np.float64)
result = (out_resized_clipped[:, :, np.newaxis] + 0.25)/1.25 * inp.astype(np.float64).astype(np.uint8)