def get(self, album_id, image_id, extension=None):
"""GET handler for GGB image metadata and files.
URL pattern: /albums/${album_id}/images/${image_id}(${extension})
If called without a file extension:
If image exists, returns 200 OK with JSON image data structure.
Returns Content-type: application/json.
If image doesn't exist, returns 404 NOT FOUND.
If called with a file extension:
If image exists and has the matching extension, returns the image.
Returned Content-type matches the image format.
Otherwise returns 404 NOT FOUND.
Returns 401 UNAUTHORIZED to all calls if authorization fails.
"""
q = Album.all().filter('album_id =', album_id)
album = q.get()
if not album:
return self.error(404)
q = Image.all().filter('album =', album).filter('image_id =', image_id)
image = q.get()
if not image:
return self.error(404)
if not extension:
data = image.to_dict()
return write_json(self, image.to_dict())
if extension != image.extension:
return self.error(404)
write_image(self, image.image_data, image.extension)
def add_face(self, name, face_image):
name = name.lower()
entry = self.find_user(name)
image_path = os.path.join('images/', f'{name}_{util.get_random_hash()}.jpg')
util.write_image(image_path, face_image)
if entry is not None:
entry['faces'].append(image_path)
else:
entry = {
'name': name,
'faces': [
image_path
]
}
self.db.append(entry)
self.flush()
def get(self, hash, extension=None):
q = Album.all().filter('hash =', hash)
album = q.get()
if album:
if extension:
return self.error(404)
q = Image.all().filter('album =', album)
return self.response.out.write(render_template('album.html', {
'name': album.name,
'images': q,
}))
q = Image.all().filter('hash =', hash)
image = q.get()
if image:
if not extension:
return self.response.out.write(render_template('image.html',
{ 'image': image }))
elif image.extension == extension:
return write_image(self, image.image_data, extension)
else:
return self.error(404)
return self.error(404)
def write_image_output(output_img, content_img, style_imgs, init_img):
out_dir = os.path.join(args.img_output_dir, args.image_name)
mkdir(out_dir)
img_path = os.path.join(out_dir, args.image_name+'.png')
content_path = os.path.join(out_dir, 'content.png')
init_path = os.path.join(out_dir, 'init.png')
write_image(img_path, output_img)
write_image(content_path, content_img)
write_image(init_path, init_img)
index = 0
for style_img in style_imgs:
path = os.path.join(out_dir, 'style_'+str(index)+'.png')
write_image(path, style_img)
index += 1
"""
Additive brightness control.
Resulting pixel value = original pixel value + c, c integer).
Watch out for R, G and B bounds!
"""
import cv2 as cv
import color
import util
BRIGHTNESS_FACTOR = 100
name = "skate.jpg"
original = util.read_image(name)
bright = color.additive_brightness(original, BRIGHTNESS_FACTOR)
cv.imshow('Additive Brightness', bright)
util.write_image("rgb-add-bright-" + name, bright)
yiq = color.rgb_to_yiq(original)
bright = color.additive_brightness(yiq, BRIGHTNESS_FACTOR)
cv.imshow('Additive Brightness', bright)
util.write_image("yiq-add-bright-" + name, bright)
cv.waitKey(0)
cv.destroyAllWindows()
def numpixels(image):
return image.view(rgb_dtype).size
def test(image):
return numcolors(image) == numpixels(image)
if __name__ == '__main__':
import argparse
import sys
parser = argparse.ArgumentParser(description="Generates an image with every RGB color exactly once")
parser.add_argument('command',
choices=['generate', 'test'],
default='generate')
parser.add_argument('-i', '--input',
type=argparse.FileType('r'),
default=sys.stdin)
parser.add_argument('-o', '--output',
type=argparse.FileType('w'),
default=sys.stdout)
args = parser.parse_args()
image = read_image(args.input)
if args.command == 'generate':
write_image(args.output, generate(image))
elif args.command == 'test':
imagecolors = numcolors(image)
imagepixels = numpixels(image)
print("This image has {} colors and {} pixels".format(imagecolors, imagepixels))
sys.exit(imagecolors != imagepixels)
# Processing
red_band = color.red_band(original)
green_band = color.green_band(original)
blue_band = color.blue_band(original)
mono_red = color.monochromatic_red(original)
mono_green = color.monochromatic_green(original)
mono_blue = color.monochromatic_blue(original)
gray = color.rgb_to_gray(original)
# Save
util.write_image("red-" + name, red_band)
util.write_image("green-" + name, green_band)
util.write_image("blue-" + name, blue_band)
util.write_image("mono-red-" + name, mono_red)
util.write_image("mono-green-" + name, mono_green)
util.write_image("mono-blue-" + name, mono_blue)
util.write_image("gray-" + name, gray)
# Show
cv.imshow('Red', red_band)
cv.imshow('Green', green_band)
cv.imshow('Blue', blue_band)
cv.imshow('Monochromatic Red', mono_red)
cv.imshow('Monochromatic Green', mono_green)
"""
Custom filters
"""
import cv2 as cv
import filter
import util
import color
name = "skate.jpg"
# name = "32bits.png"
original = util.read_image(name)
cv.imshow('Original', original)
# Custom 1
custom1 = filter.custom_filter1(original)
cv.imshow('Custom 1', custom1)
util.write_image("custom1-filter-" + name, custom1)
# Custom 2
custom2 = filter.custom_filter2(original)
cv.imshow('Custom 2', custom2)
util.write_image("custom2-filter-" + name, custom2)
cv.waitKey(0)
cv.destroyAllWindows()
"""
import cv2 as cv
import color
import util
import filter
name = 'tree.jpg'
img = util.read_image(name)
lightness = 255
# RGB to Gray
img_gray = color.rgb_to_gray(img)
cv.imshow('RGB to Gray (r=g=b)', img_gray)
util.write_image("gray-" + name, img_gray)
# Gray to Expansion
img_expansion = filter.histogram_expansion(img_gray, lightness)
cv.imshow('Histogram Expansion L=' + str(lightness), img_expansion)
util.write_image('hist-exp-l' + str(lightness) + '-' + name, img_expansion)
# Expansion to Equalization
img_equalization = filter.histogram_equalization(img_expansion)
cv.imshow('Histogram Expansion L=' + str(lightness) + ' + Equalization',
img_equalization)
util.write_image('hist-exp-l' + str(lightness) + '-eq-' + name,
img_equalization)
# Gray to Equalization
img_equalization = filter.histogram_equalization(img_gray)
def write_augmentation(example, img, car_mask, road_mask, prev_ex, augmentation_name, augmentation): base = "/tmp/output/augment/" + str(ex) + "_" img, car_mask, road_mask = augmentation(img, car_mask, road_mask, prev_ex[0], prev_ex[1]) util.write_image(base + augmentation_name + ".png", img) util.write_mask(base + augmentation_name + "_car.png", car_mask) util.write_mask(base + augmentation_name + "_road.png", road_mask) prev_ex = (None,None) for ex in examples: pre_out = "/tmp/output/preprocessing/" + str(ex) car_out = "/tmp/output/infer_car/" + str(ex) road_out = "/tmp/output/infer_road/" + str(ex) augment_out = "/tmp/output/augment/" + str(ex) img = util.read_train_image(ex) util.write_image(pre_out + ".png", img) util.write_image(car_out + ".png", img) util.write_image(road_out + ".png", img) util.write_image(augment_out + ".png", img) road_mask, car_mask = util.read_masks(ex) util.write_mask(car_out + "_truth.png",car_mask) util.write_mask(road_out + "_truth.png",road_mask) cropped = util.crop(img,util.preprocess_opts) util.write_image(pre_out + "_crop.png", cropped) uncropped = util.uncrop_image(cropped,util.preprocess_opts) util.write_image(pre_out + "_uncrop.png", uncropped) preprocessed = util.preprocess_input_image(img,util.preprocess_opts) car_infer = car_model.predict(np.array([preprocessed]), batch_size=1)[0] car_infer = util.postprocess_output(car_infer, util.preprocess_opts) util.write_probability(car_out + "_infer.png", car_infer) road_infer = road_model.predict(np.array([preprocessed]), batch_size=1)[0]
# RGB to Negative
import cv2 as cv
import color
import util
name = "skate.jpg"
original = util.read_image(name)
negative = color.negative(original)
cv.imshow('RGB Negative', negative)
util.write_image("rgb-neg-" + name, negative)
yiq = color.yiq_to_rgb(original)
yiq_negative = color.negative(yiq)
cv.imshow('YIQ Negative', yiq_negative)
util.write_image("yiq-neg-" + name, yiq_negative)
cv.waitKey(0)
cv.destroyAllWindows()
"""
Thresholding on Y with a median m.
Case 1: m from user input
Case 2: m from mean of Y values
"""
import cv2 as cv
import color
import util
THRESHOLDING_FACTOR = 127
name = "skate.jpg"
original = util.read_image(name)
# Mean from user input
black_white_user = color.thresholding_user(original, THRESHOLDING_FACTOR)
cv.imshow('Thresholding User Input', black_white_user)
util.write_image("thres-user-" + name, black_white_user)
# Mean calculated from Y component
black_white_mean = color.thresholding_mean(original)
cv.imshow('Thresholding Mean', black_white_mean)
util.write_image("thres-mean-" + name, black_white_mean)
cv.waitKey(0)
cv.destroyAllWindows()
import dct
import util
# Original
name = "lena256.jpg"
img = util.read_image(name)
# DCT
dct_img = dct.transform_2d(img)
util.write_image("dct_" + name, dct_img)
# IDCT
idct_img = dct.i_transform_2d(dct_img)
util.write_image("idct_" + name, idct_img)
# RGB-YIQ-RGB conversion
import cv2 as cv
import color
import util
# An image is a matrix with the dimensions [w][h][3]. 3 for R, G and B
name = "32bits.png"
original = util.read_image(name)
# RGB to YIQ
yiq = color.rgb_to_yiq(original)
cv.imshow('YIQ', yiq)
util.write_image("yiq-" + name, yiq)
# YIQ to RGB
rgb = color.yiq_to_rgb(yiq)
cv.imshow('RGB', rgb)
util.write_image("rgb-" + name, rgb)
cv.waitKey(0)
cv.destroyAllWindows()
def write_augmentation(example, img, car_mask, road_mask, prev_ex, augmentation_name, augmentation): base = "/tmp/output/augment/" + str(ex) + "_" img, car_mask, road_mask = augmentation(img, car_mask, road_mask, prev_ex[0], prev_ex[1]) util.write_image(base + augmentation_name + ".png", img) util.write_mask(base + augmentation_name + "_car.png", car_mask) util.write_mask(base + augmentation_name + "_road.png", road_mask)
if __name__ == '__main__':
import argparse
import sys
parser = argparse.ArgumentParser(
description="Generates an image with every RGB color exactly once")
parser.add_argument('command',
choices=['generate', 'test'],
default='generate')
parser.add_argument('-i',
'--input',
type=argparse.FileType('r'),
default=sys.stdin)
parser.add_argument('-o',
'--output',
type=argparse.FileType('w'),
default=sys.stdout)
args = parser.parse_args()
image = read_image(args.input)
if args.command == 'generate':
write_image(args.output, generate(image))
elif args.command == 'test':
imagecolors = numcolors(image)
imagepixels = numpixels(image)
print("This image has {} colors and {} pixels".format(
imagecolors, imagepixels))
sys.exit(imagecolors != imagepixels)
import cv2 as cv
import filter
import util
import color
name = "skate.jpg"
# name = "lenna.png"
original = util.read_image(name)
# cv.imshow('Original', original)
# Sobel X
# sobel_x = cv.Sobel(original, cv.CV_64F, 1, 0, 3)
sobel_x = filter.sobel_x(original)
cv.imshow('Sobel X', sobel_x)
util.write_image("sobelx-filter-" + name, sobel_x)
# Sobel Y
# sobel_y = cv.Sobel(original, cv.CV_64F, 0, 1, 3)
sobel_y = filter.sobel_y(original)
cv.imshow('Sobel Y', sobel_y)
util.write_image("sobely-filter-" + name, sobel_y)
# Sobel XY
# sobel_xy = cv.Sobel(original, cv.CV_64F, 1, 1, 3)
sobel_xy = filter.sobel_xy(original)
cv.imshow('Sobel XY', sobel_xy)
util.write_image("sobelxy-filter-" + name, sobel_xy)
# Laplace
# laplace = cv.Laplacian(original, cv.CV_64F)
def infer(
gitapp: controller.GetInputTargetAndPredictedParameters,
restore_directory: str,
output_directory: str,
extract_patch_size: int,
stitch_stride: int,
infer_size: int,
channel_whitelist: Optional[List[str]],
simplify_error_panels: bool,
):
"""Runs inference on an image.
Args:
gitapp: GetInputTargetAndPredictedParameters.
restore_directory: Where to restore the model from.
output_directory: Where to write the generated images.
extract_patch_size: The size of input to the model.
stitch_stride: The stride size when running model inference.
Equivalently, the output size of the model.
infer_size: The number of simultaneous inferences to perform in the
row and column dimensions.
For example, if this is 8, inference will be performed in 8 x 8 blocks
for a batch size of 64.
channel_whitelist: If provided, only images for the given channels will
be produced.
This can be used to create simpler error panels.
simplify_error_panels: Whether to create simplified error panels.
Raises:
ValueError: If
1) The DataParameters don't contain a ReadPNGsParameters.
2) The images must be larger than the input to the network.
3) The graph must not contain queues.
"""
rpp = gitapp.dp.io_parameters
if not isinstance(rpp, data_provider.ReadPNGsParameters):
raise ValueError(
'Data provider must contain a ReadPNGsParameter, but was: %r',
gitapp.dp)
original_crop_size = rpp.crop_size
image_num_rows, image_num_columns = util.image_size(rpp.directory)
logging.info('Uncropped image size is %d x %d', image_num_rows,
image_num_columns)
image_num_rows = min(image_num_rows, original_crop_size)
if image_num_rows < extract_patch_size:
raise ValueError(
'Image is too small for inference to be performed: %d vs %d',
image_num_rows, extract_patch_size)
image_num_columns = min(image_num_columns, original_crop_size)
if image_num_columns < extract_patch_size:
raise ValueError(
'Image is too small for inference to be performed: %d vs %d',
image_num_columns, extract_patch_size)
logging.info('After cropping, input image size is (%d, %d)',
image_num_rows, image_num_columns)
num_row_inferences = (image_num_rows -
extract_patch_size) // (stitch_stride * infer_size)
num_column_inferences = (image_num_columns - extract_patch_size) // (
stitch_stride * infer_size)
logging.info('Running %d x %d inferences', num_row_inferences,
num_column_inferences)
num_output_rows = (num_row_inferences * infer_size * stitch_stride)
num_output_columns = (num_column_inferences * infer_size * stitch_stride)
logging.info('Output image size is (%d, %d)', num_output_rows,
num_output_columns)
g = tf.Graph()
with g.as_default():
row_start = tf.placeholder(dtype=np.int32, shape=[])
column_start = tf.placeholder(dtype=np.int32, shape=[])
# Replace the parameters with a new set, which will cause the network to
# run inference in just a local region.
gitapp = gitapp._replace(dp=gitapp.dp._replace(
io_parameters=rpp._replace(
row_start=row_start,
column_start=column_start,
crop_size=(infer_size - 1) * stitch_stride +
extract_patch_size,
)))
visualization_lts = controller.setup_stitch(gitapp)
def get_statistics(tensor):
rc = lt.ReshapeCoder(list(tensor.axes.keys())[:-1], ['batch'])
return rc.decode(ops.distribution_statistics(rc.encode(tensor)))
visualize_input_lt = visualization_lts['input']
visualize_predict_input_lt = get_statistics(
visualization_lts['predict_input'])
visualize_target_lt = visualization_lts['target']
visualize_predict_target_lt = get_statistics(
visualization_lts['predict_target'])
input_lt = lt.LabeledTensor(tf.placeholder(
dtype=np.float32,
shape=[
1, num_output_rows, num_output_columns,
len(gitapp.dp.input_z_values), 1, 2
]),
axes=[
'batch',
'row',
'column',
('z', gitapp.dp.input_z_values),
('channel', ['TRANSMISSION']),
('mask', [False, True]),
])
predict_input_lt = lt.LabeledTensor(
tf.placeholder(
dtype=np.float32,
shape=[
1,
num_output_rows,
num_output_columns,
len(gitapp.dp.input_z_values),
1,
len(visualize_predict_input_lt.axes['statistic']),
]),
axes=[
'batch',
'row',
'column',
('z', gitapp.dp.input_z_values),
('channel', ['TRANSMISSION']),
visualize_predict_input_lt.axes['statistic'],
])
input_error_panel_lt = visualize.error_panel_from_statistics(
input_lt, predict_input_lt, simplify_error_panels)
target_lt = lt.LabeledTensor(
tf.placeholder(dtype=np.float32,
shape=[
1, num_output_rows, num_output_columns,
len(gitapp.dp.target_z_values),
len(gitapp.dp.target_channel_values) + 1, 2
]),
axes=[
'batch',
'row',
'column',
('z', gitapp.dp.target_z_values),
('channel',
gitapp.dp.target_channel_values + ['NEURITE_CONFOCAL']),
('mask', [False, True]),
])
predict_target_lt = lt.LabeledTensor(
tf.placeholder(
dtype=np.float32,
shape=[
1,
num_output_rows,
num_output_columns,
len(gitapp.dp.target_z_values),
len(gitapp.dp.target_channel_values) + 1,
len(visualize_predict_target_lt.axes['statistic']),
]),
axes=[
'batch',
'row',
'column',
('z', gitapp.dp.target_z_values),
('channel',
gitapp.dp.target_channel_values + ['NEURITE_CONFOCAL']),
visualize_predict_target_lt.axes['statistic'],
])
logging.info('input_lt: %r', input_lt)
logging.info('predict_input_lt: %r', predict_input_lt)
logging.info('target_lt: %r', target_lt)
logging.info('predict_target_lt: %r', predict_target_lt)
def select_channels(tensor):
if channel_whitelist is not None:
return lt.select(tensor, {'channel': channel_whitelist})
else:
return tensor
target_error_panel_lt = visualize.error_panel_from_statistics(
select_channels(target_lt), select_channels(predict_target_lt),
simplify_error_panels)
# There shouldn't be any queues in this configuration.
queue_runners = g.get_collection(tf.GraphKeys.QUEUE_RUNNERS)
if queue_runners:
raise ValueError('Graph must not have queues, but had: %r',
queue_runners)
logging.info('Attempting to find restore checkpoint in %s',
restore_directory)
init_fn = util.restore_model(restore_directory,
restore_logits=True,
restore_global_step=True)
with tf.Session() as sess:
logging.info('Generating images')
init_fn(sess)
input_rows = []
predict_input_rows = []
target_rows = []
predict_target_rows = []
for infer_row in range(num_row_inferences):
input_row = []
predict_input_row = []
target_row = []
predict_target_row = []
for infer_column in range(num_column_inferences):
rs = infer_row * infer_size * stitch_stride
cs = infer_column * infer_size * stitch_stride
logging.info('Running inference at offset: (%d, %d)', rs,
cs)
[inpt, predict_input, target,
predict_target] = sess.run([
visualize_input_lt,
visualize_predict_input_lt,
visualize_target_lt,
visualize_predict_target_lt,
],
feed_dict={
row_start: rs,
column_start: cs
})
input_row.append(inpt)
predict_input_row.append(predict_input)
target_row.append(target)
predict_target_row.append(predict_target)
input_rows.append(np.concatenate(input_row, axis=2))
predict_input_rows.append(
np.concatenate(predict_input_row, axis=2))
target_rows.append(np.concatenate(target_row, axis=2))
predict_target_rows.append(
np.concatenate(predict_target_row, axis=2))
logging.info('Stitching')
stitched_input = np.concatenate(input_rows, axis=1)
stitched_predict_input = np.concatenate(predict_input_rows, axis=1)
stitched_target = np.concatenate(target_rows, axis=1)
stitched_predict_target = np.concatenate(predict_target_rows,
axis=1)
logging.info('Creating error panels')
[input_error_panel, target_error_panel, global_step] = sess.run(
[
input_error_panel_lt, target_error_panel_lt,
tf.train.get_global_step()
],
feed_dict={
input_lt: stitched_input,
predict_input_lt: stitched_predict_input,
target_lt: stitched_target,
predict_target_lt: stitched_predict_target,
})
output_directory = os.path.join(output_directory,
'%.8d' % global_step)
if not gfile.Exists(output_directory):
gfile.MakeDirs(output_directory)
util.write_image(
os.path.join(output_directory, 'input_error_panel.png'),
input_error_panel[0, :, :, :])
util.write_image(
os.path.join(output_directory, 'target_error_panel.png'),
target_error_panel[0, :, :, :])
logging.info('Done generating images')
"""
Multiplicative brightness control.
Resulting pixel value = original pixel value * c, c integer).
Watch out for R, G and B bounds!
"""
import cv2 as cv
import color
import util
BRIGHTNESS_FACTOR = 0.25
name = "skate.jpg"
original = util.read_image(name)
# RGB Brightness
bright = color.multiplicative_brightness(original, BRIGHTNESS_FACTOR)
cv.imshow('Multiplicative Brightness', bright)
util.write_image("mult-bright-" + name, bright)
# YIQ Brightness
yiq = color.rgb_to_yiq(original)
bright = color.multiplicative_brightness(yiq, BRIGHTNESS_FACTOR)
cv.imshow('Multiplicative Brightness', bright)
util.write_image("yiq-mult-bright-" + name, bright)
cv.waitKey(0)
cv.destroyAllWindows()
[0, 0, 2],
[0, -1, 0],
[-1, 0, 0]
])
_offset = 1
# Original
name = "monkey.jpg"
_img = util.read_image(name)
cv.imshow("Original", _img)
# Gray
img_gray = color.rgb_to_gray(_img)
cv.imshow("Gray", img_gray)
util.write_image("gray-" + name, img_gray)
# Convolution Kernel c1
util.do_convolution_show_and_write(_img, kernel_c1, "conv-c1-" + name, _offset)
# Convolution Gray Kernel c1
util.do_convolution_show_and_write(img_gray, kernel_c1, "conv-c1-gray-" + name, _offset)
# 2
util.do_convolution_show_and_write(_img, kernel_c2, "conv-c2-" + name, _offset)
util.do_convolution_show_and_write(img_gray, kernel_c2, "conv-c2-gray-" + name, _offset)
# 3
util.do_convolution_show_and_write(_img, kernel_c3, "conv-c3-" + name, _offset)
util.do_convolution_show_and_write(img_gray, kernel_c3, "conv-c3-gray-" + name, _offset)
import cv2 as cv
import util
# Equalization with OpenCV
img_cv = cv.imread('img/tree.jpg', 0)
cv_eq = cv.equalizeHist(img_cv)
cv.imshow('cv-eq-tree.jpg', cv_eq)
util.write_image('cv-eq-tree.jpg', cv_eq)
cv.waitKey(0)
cv.destroyAllWindows()
