def prepare_img(img_path, model_name='InceptionV3'):
img = image.load_img(img_path, target_size=target_sizes[model_name])
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
if model_name == 'InceptionV3':
return inception_v3_preprocess(x)
if model_name == 'ResNet50':
return resnet50_preprocess(x)
if model_name == 'VGG16':
return vgg16_preprocess(x)
def preprocess_vgg16(img_path):
"""
Loads image and applies preprocessing based on VGG16-models and returns 4D-array.
Parameters
----------
img_path: str
pth to image
Returns
-------
np.array
"""
return vgg16_preprocess(path_to_tensor(img_path, scale=False))
def get_image_value(path, dim, bw, model_type):
'''This function will read an image and convert to a specified version and resize depending on which algorithm is being used. If edge is specified as true, it will pass the img array to get_edged which returns a filtered version of the img'''
img = image.load_img(path, target_size=dim)
img = image.img_to_array(img)
if bw == True:
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
if model_type.upper() != 'Normal':
img = np.stack((img, ) * 3, axis=-1)
else:
img = img.reshape(img.shape[0], img.shape[1], 1)
if model_type.upper() == 'MOBILENET':
img = mobile_preprocess(img)
return img
elif model_type.upper() == 'VGG16':
img = vgg16_preprocess(img)
return img
return img / 255
movie_list = [subdir for subdir in os.listdir(jpg_dir)]
vgg16_model = VGG16(weights='imagenet')
vgg16_layer = Model(inputs=vgg16_model.input, outputs=vgg16_model.get_layer('fc2').output)
inceptionv3_model = InceptionV3(weights='imagenet')
inceptionv3_layer = Model(inputs=inceptionv3_model.input, outputs=inceptionv3_model.get_layer('avg_pool').output)
resnet50_model = ResNet50(weights='imagenet')
resnet50_layer = Model(inputs=resnet50_model.input, outputs=resnet50_model.get_layer('avg_pool').output)
for movie in movie_list:
clip_list = [clip for clip in os.listdir(os.path.join(jpg_dir, movie))]
for clip in clip_list:
print "Processing %s" % clip
jpg_list = [jpg for jpg in os.listdir(os.path.join(jpg_dir, movie, clip))]
vgg16_list = []
inceptionv3_list = []
resnet50_list = []
for jpg in sorted(jpg_list):
jpg_path = os.path.join(jpg_dir, movie, clip, jpg)
keras_image_224, keras_image_299 = kerasImage.load_img(jpg_path,target_size=(224, 224)), kerasImage.load_img(jpg_path, target_size=(299, 299))
keras_image_arr_224, keras_image_arr_299 = kerasImage.img_to_array(keras_image_224), kerasImage.img_to_array(keras_image_299)
keras_image_arr_224, keras_image_arr_299 = np.expand_dims(keras_image_arr_224, axis=0), np.expand_dims(keras_image_arr_299, axis=0)
vgg16_list.append(vgg16_layer.predict(vgg16_preprocess(keras_image_arr_224)))
inceptionv3_list.append(inceptionv3_layer.predict(inceptionv3_preprocess(keras_image_arr_299)))
resnet50_list.append(resnet50_layer.predict(resnet50_preprocess(keras_image_arr_224)))
np.save(os.path.join(jpg_dir, movie, clip+"_vgg16.npy"), np.array(vgg16_list))
np.save(os.path.join(jpg_dir, movie, clip+"_inceptionv3.npy"), np.array(inceptionv3_list))
np.save(os.path.join(jpg_dir, movie, clip+"_resnet50.npy"), np.array(resnet50_list))
from VGG16FeatureGenerator import VGG16FeatureGenerator
from keras.applications.vgg16 import VGG16, preprocess_input as vgg16_preprocess
from keras.preprocessing import image
import matplotlib.pyplot as plt
import numpy as np
if __name__ == '__main__':
img_path = sys.argv[1]
if not img_path:
exit(1)
model = VGG16(weights='imagenet', include_top=False)
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = vgg16_preprocess(x)
features = model.predict(x)
print features.shape
features = np.squeeze(features, axis=0)
print features
print features.shape
print features[:, :, -1]
print features[:, :, -1].shape
plt.imshow(features[:, :, -1])
def extract_VGG16(tensor) -> np.ndarray: return VGG16(**args_NN).predict(vgg16_preprocess(tensor)) def extract_VGG19(tensor) -> np.ndarray: return VGG19(**args_NN).predict(vgg19_preprocess(tensor))