def execute(file, curr, rank):
# Load sample image
# file = utils.DATA_DIR + 'sample.jpg'
img = image.load_img(file)
# Resize
scale = utils.IMG_SIZE / max(img.size)
new_size = (int(np.ceil(scale * img.size[0])),
int(np.ceil(scale * img.size[1])))
print(time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(time.time())),
str(curr), "is processing...")
img = img.resize(new_size)
# Mean substraction
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = utils.preprocess_image(x)
# Load RMAC model
Wmap, Hmap = get_size_vgg_feat_map(x.shape[3], x.shape[2])
regions = rmac_regions(Wmap, Hmap, 3)
# print('Loading RMAC model...')
model = rmac((x.shape[1], x.shape[2], x.shape[3]), len(regions), rank)
# Compute RMAC vector
# print('Extracting RMAC from image...')
RMAC = model.predict([x, np.expand_dims(regions, axis=0)])
return RMAC
def load_model(x, multi='parallel'):
x = np.array(x)
print('cur batch tensor shape', x.shape)
x = utils.preprocess_image(x)
# Load RMAC model
Wmap, Hmap = get_size_vgg_feat_map(x.shape[2],
x.shape[1]) #image_dim_ordering tf
#Wmap, Hmap = get_size_vgg_feat_map(x.shape[3], x.shape[2]) #image_dim_ordering th
regions = rmac_regions(Wmap, Hmap, 3)
print('Loading RMAC model...')
print(x.shape[1], x.shape[2], x.shape[3], len(regions))
if PARALLEL:
if multi == 'single':
model = rmac((x.shape[1], x.shape[2], x.shape[3]), len(regions))
return model, regions
else:
with tf.device('/cpu:0'):
model = rmac((x.shape[1], x.shape[2], x.shape[3]),
len(regions))
parallel_model = multi_gpu_model(model, gpus=num_gpu)
return parallel_model, regions
else:
model = rmac((x.shape[1], x.shape[2], x.shape[3]), len(regions))
return model, regions
def ComputeRMAC(x):
Wmap, Hmap = get_size_vgg_feat_map(
x.shape[2], x.shape[1]) #Wmap, Hmap means extract from a root 6 times
regions = rmac_regions(Wmap, Hmap, 10)
# Load RMAC model
model = rmac((x.shape[1], x.shape[2], x.shape[3]), len(regions))
# Compute RMAC vector
RMAC = model.predict([x, np.expand_dims(regions, axis=0)])
print(RMAC)
RMAC = np.reshape(RMAC, (-1, 512))
print(RMAC.shape)
return RMAC
def rmac_holidays(img):
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
# Load RMAC model
Wmap, Hmap = get_size_vgg_feat_map(x.shape[1], x.shape[2])
# warning can be ignored
regions = rmac_regions(Wmap, Hmap, 3)
print('Loading RMAC model...')
model = rmac((x.shape[1], x.shape[2], x.shape[3]), len(regions))
# Compute RMAC vector
print('Extracting RMAC from image...')
RMAC1 = model.predict([x, np.expand_dims(regions, axis=0)])
return RMAC1
def describe(img):
K.clear_session()
# Resize
scale = utils.IMG_SIZE / max(img.size)
new_size = (int(np.ceil(scale * img.size[0])),
int(np.ceil(scale * img.size[1])))
img = img.resize(new_size)
# Mean substraction
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = utils.preprocess_image(x)
# Load RMAC model
Wmap, Hmap = get_size_vgg_feat_map(x.shape[3], x.shape[2])
regions = rmac_regions(Wmap, Hmap, 3)
model = rmac((x.shape[1], x.shape[2], x.shape[3]), len(regions))
# Compute RMAC vector
RMAC = model.predict([x, np.expand_dims(regions, axis=0)])
return RMAC
# plt.draw()
# plt.pause(5)
# plt.close()
# Mean substraction
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = utils.preprocess_image(x)
# Show preprocessed image
# y = np.squeeze(x, axis=0)
# img = image.array_to_img(y)
# plt.imshow(img)
# plt.draw()
# plt.pause(5)
# plt.close()
# Load RMAC model
Wmap, Hmap = get_size_vgg_feat_map(x.shape[3], x.shape[2])
regions = rmac_regions(Wmap, Hmap, 3)
print('Loading RMAC model...')
model = rmac((x.shape[1], x.shape[2], x.shape[3]), len(regions))
# Compute RMAC vector
print('Extracting RMAC from image...')
RMAC = model.predict([x, np.expand_dims(regions, axis=0)])
print('RMAC size: %s' % RMAC.shape[1])
print(RMAC)
print('Done!')
def infer(queries, _):
test_path = DATASET_PATH + '/test/test_data'
references = [
os.path.join(test_path, 'reference', path)
for path in os.listdir(os.path.join(test_path, 'reference'))
]
queries = [v.split('/')[-1].split('.')[0] for v in queries]
references = [v.split('/')[-1].split('.')[0] for v in references]
queries.sort()
references.sort()
# Load RMAC model
Wmap, Hmap = get_size_vgg_feat_map(224, 224)
regions = rmac_regions(Wmap, Hmap, 3)
rmac_model = rmac(model, len(regions))
rmac_model.summary()
test_datagen = ImageDataGenerator(dtype='float32')
query_generator = test_datagen.flow_from_directory(directory=test_path,
target_size=(224,
224),
classes=['query'],
color_mode="rgb",
batch_size=1,
class_mode=None,
shuffle=False)
query_vecs = rmac_model.predict_generator(multi_input(
query_generator, regions),
steps=len(query_generator),
verbose=1)
reference_generator = test_datagen.flow_from_directory(
directory=test_path,
target_size=(224, 224),
classes=['reference'],
color_mode="rgb",
batch_size=1,
class_mode=None,
shuffle=False)
reference_vecs = rmac_model.predict_generator(
multi_input(reference_generator, regions),
steps=len(reference_generator),
verbose=1)
# l2 normalization
query_vecs = l2_normalize(query_vecs)
reference_vecs = l2_normalize(reference_vecs)
# Calculate cosine similarity
sim_matrix = np.dot(query_vecs, reference_vecs.T)
indices = np.argsort(sim_matrix, axis=1)
indices = np.flip(indices, axis=1)
retrieval_results = {}
for (i, query) in enumerate(queries):
ranked_list = [references[k] for k in indices[i]]
ranked_list = ranked_list[:1000]
retrieval_results[query] = ranked_list
print('done')
return list(
zip(range(len(retrieval_results)), retrieval_results.items()))