def get_photos_from_url(gender, type, url):
crawler.open(url)
crawler.infinite_scroll(slowly=True)
images = crawler.find_elements('img.product-card--img')
srcs = img2src(images)
print '{} images got'.format(len(images))
download_images(srcs, os.path.join('gap', gender, type))
def get_photos_from_url(gender, type, url):
crawler.open(url)
crawler.infinite_scroll(slowly=True)
images = crawler.find_elements('img.photo.default')
srcs = img2src(images)
print '{} images got'.format(len(images))
download_images(srcs, os.path.join('armanicollezioni', gender, type))
def main():
parser = argparse.ArgumentParser(description='Totality image classifier using VGG in Keras')
parser.add_argument('--download', default=False, action='store_true')
parser.add_argument('--img-dir', type=str, default='img')
parser.add_argument('--label-file', type=str, default='labeled_data.json')
parser.add_argument('--output-file', type=str, default='vgg_output_labels.json')
args = parser.parse_args()
if args.download:
with open(args.label_file) as f:
download_images(args.img_dir, list(json.load(f).keys()))
print('Collecting dataset...')
dataset = ImageDataNP(args.img_dir, labeled_data_file=args.label_file, train_ratio=1.0, squash=True)
x_train, y_train = dataset.get_train()
model = get_vgg('VGG16', weights='imagenet', classes=1000)
writer = PredictionWriter()
batch_size=32
start, end = 0, 0
print('Classifying images...')
for i in range(int((x_train.shape[0] + batch_size - 1) / batch_size)):
end = min((i + 1) * batch_size, x_train.shape[0])
print('Batch {}, images[{}:{}]'.format(i, start, end))
# Obtain predictions and print results
preds = model.predict(x_train[start:end])
record_batch_results(writer, preds, y_train[start:end], dataset.train_keys, start)
# Update start index
start = end
writer.commit(args.output_file)
def get_photos_from_url(gender, type, url):
crawler.open(url)
images = crawler.find_elements('span.back' '> img')
srcs = img2src(images)
print '{} images got'.format(len(images))
download_images(srcs, os.path.join('kuho', gender, type))
def download_nam3km_soundings(run, lat, lon):
for fh in range(NAM3KM_SOUNDING_HOURS):
sounding_url = sounding_img_url("nam3km", run, fh, lat, lon)
sounding_filepath = "nam3km_sounding_fh" + str(fh).zfill(2) + ".png"
logger.info("Downloading sounding: {:s} -> {:s}".format(sounding_url, sounding_filepath))
download_images(sounding_url, filename=sounding_filepath)
def get_photos_from_url(gender, type, url):
crawler.open(url)
crawler.infinite_scroll(slowly=True)
images = crawler.find_elements('div.item-imgwrap' '> img:not(.hover)')
srcs = img2src(images)
print '{} images got'.format(len(images))
download_images(srcs, os.path.join('guess', gender, type))
def get_photos_from_url(gender, type, url):
crawler.open(url)
crawler.infinite_scroll(sleep_time=3.5)
images = crawler.find_elements('img.product-tile__image')
srcs = img2src(images)
print '{} images got'.format(len(images))
download_images(srcs, os.path.join('hugoboss', gender, type))
def get_photos_from_url(gender, type, url):
crawler.open(url)
crawler.infinite_scroll()
images = crawler.find_elements('img.grid-product__image')
srcs = img2src(images)
print '{} images got'.format(len(images))
if gender == 'women':
srcs = map(lambda src: src.replace('model', 'prod'), srcs)
download_images(srcs, os.path.join('hollister', gender, type))
def get_photos_from_url(gender, type, url):
crawler.open(url)
images = crawler.find_elements(
'a.product-tiles-grid-item-link '
'> div.product-tiles-grid-item-image-wrapper '
'> div.product-tiles-grid-item-image '
'> img')
srcs = img2src(images)
print '{} images got'.format(len(images))
download_images(srcs, os.path.join('gucci', gender, type))
def get_photos_from_url(gender, type, url):
crawler.open(url)
crawler.infinite_scroll(sleep_time=3.)
images = crawler.find_elements('figure > noscript')
def parse_src(elem):
html = elem.get_attribute('innerHTML')
for token in html.split(' '):
if 'src' in token:
src = token.split('"')[-2]
return 'http://www.cosstores.com{}'.format(src)
srcs = map(parse_src, images)
print '{} images got'.format(len(images))
download_images(srcs, os.path.join('cos', gender, type))
def main():
parser = argparse.ArgumentParser(description='A test')
parser.add_argument('--download', default=False, action='store_true')
parser.add_argument('--img-dir', type=str, default='img')
parser.add_argument('--label-file', type=str, default='labeled_data.json')
parser.add_argument('--save', type=str, default=None)
args = parser.parse_args()
if args.download:
with open(args.label_file) as f:
download_images(args.img_dir, list(json.load(f).keys()))
# VGG Phase
print('Collecting dataset...')
dataset = ImageDataNP(args.img_dir,
labeled_data_file=args.label_file,
train_ratio=1.0,
squash=True)
x_train, y_train = dataset.get_train()
vgg = get_vgg('VGG19', weights='imagenet', classes=1000)
inp = vgg.input
out = vgg.get_layer(name='fc2').output
functor = K.function([inp], [out])
vgg_relu_out = list()
batch_size = 32
start, end = 0, 0
print('Classifying images...')
for i in range(int((x_train.shape[0] + batch_size - 1) / batch_size)):
end = min((i + 1) * batch_size, x_train.shape[0])
print('Batch {}, images[{}:{}]'.format(i, start, end))
vecs = functor([x_train[start:end]])
vgg_relu_out.extend(vecs[0])
# Update start index
start = end
# Logistic Regression Phase
kfold_dataset = ImageDataKFold(nfolds=10,
custom_data=(np.array(vgg_relu_out),
y_train))
model = train_and_eval_lr(kfold_dataset)
# Save model?
if args.save is not None:
model.save(args.save)
def get_photos_from_url(gender, type, url):
crawler.open(url)
button = crawler.find_element('button.cookie-compliance-ok-btn')
if button is not None:
button.click()
crawler.infinite_scroll()
if gender == 'men':
images = crawler.find_elements('img.product-image-front.active')
else:
images = crawler.find_elements('img.product-image-front.active')
# images = crawler.find_elements('img.product-image-reverse')
srcs = img2srcset(images)
srcs = filter(lambda src: src.strip() != '', srcs)
srcs = map(
lambda src: 'http:{}'.format(
src.split(',')[-1].strip().split(' ')[0]), srcs)
print '{} images got'.format(len(srcs))
download_images(srcs, os.path.join('americaneagle', gender, type))
def index():
payload = {
'api_key': config.LASTFM_API_KEY,
'method': 'chart.getTopArtists',
'limit': 9,
'format': 'json'
}
r = requests.get(config.url, params=payload)
links = gen_artistlinks(r.text)
filenames = download_images(links, 'index')
generate_collage(filenames, 'index_image')
return render_template('index.html')
def main():
parser = argparse.ArgumentParser(description='Totality image classifier')
parser.add_argument('--download', default=False, action='store_true')
parser.add_argument('--img-dir', type=str, default='img')
parser.add_argument('--img-uri', type=str, default='img/img.txt')
parser.add_argument('--lr-model', type=str, default='lr_model.h5')
parser.add_argument('--save', type=str, default=None)
args = parser.parse_args()
# Create URI list
with open(args.img_uri) as f:
img_uri = [l.strip() for l in f.readlines()]
if args.download:
download_images(args.img_dir, img_uri)
# Create list of paths to local image files
img_paths = [
os.path.join(args.img_dir, os.path.basename(i)) for i in img_uri
]
# ImageSet object will provide images in for loop below.
# This object provides a generator to access batches of
# images, allowing only one batch to be loaded into RAM
# at a time.
images = ImageSet(img_paths)
# Create VGG19 model
vgg = get_vgg('VGG19', weights='imagenet', classes=1000)
inp = vgg.input
out = vgg.get_layer(name='fc2').output
functor = K.function([inp], [out])
print('Classifying images...')
vgg_fc2_out = list()
batch_size = 32
i = 0
for batch, start, end in images.get_batches(batch_size):
print('Batch {}, images[{}:{}]'.format(i, start, end))
vecs = functor([batch])
vgg_fc2_out.extend(vecs[0])
i += 1
# Load trained logistic regression model
model = load_model(args.lr_model)
# Classify tensors from VGG19 using LR model
predictions = model.predict(np.array(vgg_fc2_out))
# Prediction dictionary. Used if the --save argument is supplied
pred_dict = dict()
# Print predictions and assemble pred_dict
for i in range(len(img_uri)):
cur_img_uri = img_uri[i]
cls_name = decode_totality_prediction(predictions[i])
pred_dict[cur_img_uri] = cls_name
print(cur_img_uri, cls_name)
# Save predictions?
if args.save is not None:
with open(args.save, 'w') as f:
json.dump(pred_dict, f)
def get_product_images(dir):
crawler.infinite_scroll()
images = crawler.find_elements('article.product-item > a > img')
print '{} images got'.format(len(images))
srcs = img2src(images)
download_images(srcs, os.path.join('hnm', gender, type, dir))
import utils
tf.enable_eager_execution()
assert tf.__version__ == '1.13.1' or tf.__version__ == '1.14.0', "This code is only tested on Tensorflow 1.13.1 and 1.14.0"
# Define model constants.
flags = tf.app.flags
flags = utils.create_flags(flags)
# Use third party images with 102 categories flowers.
BATCH_SIZE = 128
EPOCH = 7
INPUT_SIZE = 8189
BUFFER_SIZE = 8000
NUM_CLASSES = 102
iter_number = (int)(INPUT_SIZE / BATCH_SIZE) + 1
image_root = utils.download_images()
train_ds = utils.load_data(image_root)
train_ds = utils.prepare_train_ds(train_ds,
BATCH_SIZE,
BUFFER_SIZE,
image_size=128)
# Guild generator and discriminator model.
generator_net = Generator(dtype=flags.FLAGS.dtype)
generator_optimizer = tf.train.AdamOptimizer(
learning_rate=flags.FLAGS.learning_rate_generator,
beta1=flags.FLAGS.beta1,
beta2=flags.FLAGS.beta2)
discriminator_net = Discriminator(alpha=flags.FLAGS.alpha,
dtype=flags.FLAGS.dtype)
def init_tests():
"""Get the data required for the tests."""
get_cifar10_dataset()
convert_cifar2imagenet_format()
download_images()
get_models()
