def save_generated_images(generated_images, generator_iteration):
# Create the plot
plt.figure(figsize=(8, 8), num=1)
gs1 = gridspec.GridSpec(8, 8)
gs1.update(wspace=0, hspace=0)
for i in range(64):
ax1 = plt.subplot(gs1[i])
ax1.set_aspect('equal')
image = generated_images[i, :, :, :]
image += 1
image *= 127.5
fig = plt.imshow(image.astype(np.uint8))
plt.axis('off')
fig.axes.get_xaxis().set_visible(False)
fig.axes.get_yaxis().set_visible(False)
plt.tight_layout()
save_name = 'generated images/generatedSamples_genIter' + \
str(generator_iteration + 1) + '.png'
plt.savefig(save_name, bbox_inches='tight', pad_inches=0)
plt.pause(0.0000000001)
plt.show()
def plot_img_with_class_activation_map(image, class_activation_map):
image_height = image.shape[0]
image_width = image.shape[1]
cam_resized = tf.image.resize(class_activation_map, [image_height, image_width]).numpy()
fig, ax = plt.subplots()
ax.imshow(image.astype(int))
ax.imshow(cam_resized, cmap='jet', alpha=0.5)
plt.axis('off')
return fig
def HistEqualization(image, number_bins = 256): #implementing the histogram equalization
# get the image histogram
image_Hist, bins = np.histogram(image.flatten(), number_bins, [0, 256])
cdf = image_Hist.cumsum() # cumulative distribution function
cdf = image_Hist.max()*cdf/cdf.max() #normalize
cdf_mask = np.ma.masked_equal(cdf, 0)
cdf_mask = (cdf_mask - cdf_mask.min())*255/(cdf_mask.max()-cdf_mask.min())
cdf = np.ma.filled(cdf_mask,0).astype('uint8')
return cdf[image.astype('uint8')]
def load_image_pixels(filename, shape):
image = load_img(filename, target_size=shape)
# convert to numpy array
image = img_to_array(image)
# scale pixel values to [0, 1]
image = image.astype('float32')
image /= 255.0
# add a dimension so that we have one sample
image = expand_dims(image, 0)
return image
def preprocess_image(image, target_size):
'''function to preprocess input image'''
if image.mode != "RGB":
image = image.convert("RGB") # if image is not RGB then convert to RGB
image = image.resize(target_size) # resize image to defined target size
# image = image.img_to_array(image) # convert image to an array of numbers
image = asarray(image)
image = image.astype(np.float32) # convert from uint8 to float32
image = np.expand_dims(image, axis=0) # expand the dimension of the image
return image
def traitement(image_data, model):
size = (165,133)
image = ImageOps.fit(image_data, size, Image.ANTIALIAS)
image=np.asarray(image)
image = (image.astype(np.float32) / 255.0)
img_reshape = np.expand_dims(image, axis = 0)
img_reshape = image[np.newaxis,...,np.newaxis]
prediction = model.predict(img_reshape)
return prediction
def generate_with_pillow(models, seeds, color_channels, sample=False):
if sample:
images = sample_images()
else:
model = keras.models.load_model(models[-1]) # last models is the best
images = model(seeds, training=False)
img_size = (SIZE * NUM_HEIGHT, SIZE * NUM_WIDTH)
if color_channels == 3:
img = Image.new('RGB', img_size)
for i in range(NUM_HEIGHT):
for j in range(NUM_WIDTH):
pos = i * NUM_HEIGHT + j
image = images[pos]
image = np.asarray(image)
image = image[:, :, :] * 127.5 + 127.5
# image = image[:, :, :] * 127.5 + 127.5
# print(image.astype(int))
# print(image.shape)
# image = image.reshape((SIZE, SIZE))
image = image.astype(np.uint8)
im = Image.fromarray(image)
img.paste(im, (i * SIZE, j * SIZE))
else:
img = Image.new('L', img_size)
for i in range(NUM_HEIGHT):
for j in range(NUM_WIDTH):
pos = i * NUM_HEIGHT + j
image = images[pos]
image = np.asarray(image)
image = image[:, :, 0] * 127.5 + 127.5
image = image.astype(np.uint8)
im = Image.fromarray(image)
img.paste(im, (i * SIZE, j * SIZE))
img.save('../report/collage_{}x{}_{}.png'.format(NUM_HEIGHT, NUM_WIDTH,
FOLDER_NAME))
def load_image_pixels(filename, shape):
# load the image to get its shape
image = load_img(filename)
width, height = image.size
# load the image with the required size
image = load_img(filename, target_size=shape)
# convert to numpy array
image = img_to_array(image)
# scale pixel values to [0, 1]
image = image.astype('float32')
image /= 255.0
# add a dimension so that we have one sample
image = expand_dims(image, 0)
return image, width, height
def face_alignment(image): # implementing face alignment
image = np.array(image)
image = image.astype(np.uint8)
gray_image = image
#gray_image = cv2.cvtColor(image ,cv2.COLOR_BGR2GRAY) # convert color image to grayscale image
rects = detector(gray_image ,1) # detect faces in the grayscale image
if len(rects) > 0:
images = []
for (i, rect) in enumerate(rects):
shape = predictor(image, rect)
shape = shape_to_np(shape)
rotated_image = rotate_image(image , shape)
images.append(rotated_image)
if len(rects) == 1 :
return rotated_image
else:
return images
else:
#print("Error : number of detected face is zero, so we just return original image")
return image