def predict_images(net,
images,
n_frames=10,
checkpoint='final_pixelate.ckpt',
size=300):
with tf.Session(graph=net['graph']) as sess:
saver = tf.train.Saver()
saver.restore(sess, 'final_pixelate.ckpt')
combined = []
for img_i, img in enumerate(images):
actual = img.copy()
processed = np.reshape(process(preprocess_img(img)), (100, 100))
img = utils.imcrop_tosquare(actual)
img = resize(img, (size, size))
img = np.dot(img[..., :3], [0.299, 0.587, 0.114]) / 255.0
Xs = []
for frame_i in range(n_frames):
combined.append(img)
for frame_i in range(n_frames):
feed_dict = {}
feed_dict[net['X']] = processed
preds = sess.run([net['Y_pred']], feed_dict=feed_dict)
guessed_img = resize(deprocess(preds), (size, size))
combined.append(guessed_img)
return combined
def get_celeb_images():
print('##Start get_celeb_images##')
dirname = "C:\\Users\\rahmanim\\Documents\\1TF-Gits\\ImagesFromTheInternet"
filenames = [os.path.join(dirname, fname) for fname in os.listdir(dirname)]
# Make sure we have exactly 100 image files!
filenames = filenames[:100]
assert (len(filenames) == 100)
#print(filenames)
# Read every filename as an RGB image
imgs = [plt.imread(fname)[..., :3] for fname in filenames]
# Crop every image to a square
imgs = [utils.imcrop_tosquare(img_i) for img_i in imgs]
# Then resize the square image to 100 x 100 pixels; mode='reflect'
imgs = [resize(img_i, (100, 100), mode='reflect') for img_i in imgs]
# Finally make our list of 3-D images a 4-D array with the first dimension the number of images:
imgs = np.array(imgs).astype(np.float32)
# Plot the resulting dataset:
# Make sure you "run" this cell after you create your `imgs` variable as a 4-D array!
# Make sure we have a 100 x 100 x 100 x 3 dimension array
assert (imgs.shape == (100, 100, 100, 3))
plt.figure(figsize=(10, 10))
plt.imshow(utils.montage(imgs, saveto='dataset.png'))
plt.title('Dataset of 100 images')
#plt.show()
return imgs
def preprocess_img(img):
with warnings.catch_warnings():
img = utils.imcrop_tosquare(img)
img = resize(img, (100, 100))
img = img_as_ubyte(img)
full = np.full([100, 100, 3], 255, dtype='int64')
img = full - img
img = np.dot(img[..., :3], [0.299, 0.587, 0.114])
img /= 255.0
channels = 10
new_img = []
for row_i, row in enumerate(img):
new_img.append([])
for idx, pixel in enumerate(row):
new_img[row_i].append(1 - (round(pixel * channels) / channels))
return new_img
def test_crop():
assert (utils.imcrop_tosquare(np.zeros((64, 23))).shape == (23, 23))
assert (utils.imcrop_tosquare(np.zeros((23, 53))).shape == (23, 23))
assert (utils.imcrop_tosquare(np.zeros((23, 23))).shape == (23, 23))
assert (utils.imcrop_tosquare(np.zeros((24, 53))).shape == (24, 24))
filenames = image_filenames
# Read every filename as an RGB image
filenames = [
os.path.join(foldername, fname) for fname in os.listdir(foldername)
if not os.path.isdir(fname) and 'train' in fname
]
imgs = [plt.imread(fname)[..., :3] for fname in filenames]
# Check images: need to have all the same number of channels
# in shape (W, H, C) C must be the same for all images
[print(img_index, ', ', img.shape) for img_index, img in enumerate(imgs)]
from libs import utils
# Crop every image to a square
imgs = [utils.imcrop_tosquare(img_i) for img_i in imgs]
# Then resize the square image to 100 x 100 pixels
from skimage.transform import resize
imgs = [resize(img_i, (100, 100)) for img_i in imgs]
# Finally make our list of 3-D images a 4-D array with the first dimension the number of images:
import numpy as np
imgs = np.array(imgs).astype(np.float32)
# Plot the resulting dataset:
# Make sure you "run" this cell after you create your `imgs` variable as a 4-D array!
# Make sure we have a 100 x 100 x 100 x 3 dimension array
assert (imgs.shape == (100, 100, 100, 3))
plt.figure(figsize=(10, 10))
plt.imshow(utils.montage(imgs, saveto='dataset.png'))
get_ipython().magic('ls')
plt.style.use('ggplot')
from libs import utils
utils.remove_duplicates('../daisy/')
#example of reload
import importlib
importlib.reload(utils)
import os
files = [f for f in os.listdir('../daisy') if '.jpg' in f]
len(files)
images = [plt.imread('../daisy/' + f) for f in files]
images_sq = [utils.imcrop_tosquare(i) for i in images]
min_img_sq = min(images_sq, key = lambda im: im.shape[1])
min_img_sq.shape
plt.imshow(min_img_sq[:,:,0], cmap='gray')
plt.imshow(min_img_sq[:,:,1], cmap='gray')
plt.imshow(min_img_sq[:,:,2], cmap='gray')
from scipy.misc import imresize
images_min = [imresize(im, (177, 177)) for im in images_sq]
data_min = np.array(images_min)
data_min.shape
mean_daisy = np.mean(data_min, axis=0).astype(np.uint8)
def test_crop():
assert(utils.imcrop_tosquare(np.zeros((64, 23))).shape == (23, 23))
assert(utils.imcrop_tosquare(np.zeros((23, 53))).shape == (23, 23))
assert(utils.imcrop_tosquare(np.zeros((23, 23))).shape == (23, 23))
assert(utils.imcrop_tosquare(np.zeros((24, 53))).shape == (24, 24))