def pipeline(directory):
pipe = ImagePipeline(directory)
pipe.read(sub_dirs=('casual', 'dress'))
pipe.resize((300, 300, 3))
#pipe.transform(rgb2gray, {})
pipe.vectorize()
return pipe
def pipeline(self):
pipe = ImagePipeline(self.image_path)
pipe.read(sub_dirs=('casual', 'dress'))
pipe.resize((300, 300, 3))
pipe.transform(rgb2gray, {})
pipe.vectorize()
return pipe
def pipeline(directory):
pipe = ImagePipeline(directory)
pipe.read(sub_dirs=('casual', 'dress'))
pipe.resize((300, 300, 3))
#pipe.transform(rgb2gray, {})
pipe.vectorize()
return pipe
def image_featurizer(category, sub_dir='white', edge=False, pca=True):
'''
Vectorize images (with white backgrounds) within the category.
INPUT:
category: string
sub_dir: string
edge: boolean
pca: boolean
OUTPUT:
feature_dict: dictionary
key: path (filename)
value: vectorized image
scaler: trained StandardScaler
pca_model: trained PCA model
'''
# Use ImagePipeline to read and transform the images:
base_path = 'wayfair/images/' + category + '/'
image_pipe = ImagePipeline(base_path)
image_pipe.read(sub_dirs=(sub_dir,))
image_pipe.resize((150,150,3))
image_pipe.transform(skimage.color.rgb2gray, {})
if edge:
image_pipe.transform(skimage.feature.canny, {})
image_pipe.vectorize()
features = image_pipe.features
# Do PCA if pca is passed in as true:
if pca:
scaler = StandardScaler()
features_scaled = scaler.fit_transform(features)
pca_model = PCA(n_components=100)
pca_data = pca_model.fit_transform(features_scaled)
features = pca_data
# Organize vectorized images into the dictionary (feature_dict):
paths = os.listdir(base_path + sub_dir)
paths = [x for x in paths if x[0] != '.']
feature_dict = {}
for i in xrange(len(paths)):
path = paths[i]
feature = features[i]
feature_dict[path] = feature
if pca:
return feature_dict, scaler, pca_model
else:
return feature_dict