model.compile(loss='categorical_crossentropy',
optimizer=opt,
metrics=[get_categorical_accuracy_keras])
# model.compile(loss='categorical_crossentropy',
# optimizer=opt,
# metrics=['accuracy'])
""" Load data """
print('dataset path', DATASET_PATH)
output_path = ['./img_list.pkl', './label_list.pkl']
train_dataset_path = DATASET_PATH + '/train/train_data'
print('train_dataset_path', train_dataset_path)
if nsml.IS_ON_NSML:
# Caching file
nsml.cache(train_data_loader,
data_path=train_dataset_path,
img_size=input_shape[:2],
output_path=output_path)
else:
# local에서 실험할경우 dataset의 local-path 를 입력해주세요.
train_data_loader(train_dataset_path,
input_shape[:2],
output_path=output_path)
with open(output_path[0], 'rb') as img_f:
img_list = pickle.load(img_f)
with open(output_path[1], 'rb') as label_f:
label_list = pickle.load(label_f)
# add grayscale
gray = []
for j in range(0, len(img_list)):
# model
model = Model(inputs=model.layers[0].input,
outputs=model.get_layer('block5_pool').output)
# parameter
input_shape = (224, 224, 3) # input image shape
output_path = ['mac_list_v1.pkl', 'label_info_v1.pkl']
""" Load data """
print('dataset path', DATASET_PATH)
train_dataset_path = DATASET_PATH + '/train/train_data'
nsml.cache(data_loader.train_data_mac_loader,
data_path=train_dataset_path,
img_size=input_shape,
output_path=output_path,
model=model)
with open(output_path[0], 'rb') as img_f:
img_vecs = pickle.load(img_f)
with open(output_path[1], 'rb') as label_f:
label_info = pickle.load(label_f)
# l2 norm
img_vecs = img_vecs / np.linalg.norm(img_vecs, axis=1).reshape(-1, 1)
# similarity
sim = util.cal_cos_sim(img_vecs, img_vecs)
# label_bound
if config.pause:
nsml.paused(scope=locals())
bTrainmode = False
if config.mode == 'train':
bTrainmode = True
""" Load data """
print(DATASET_PATH)
output_path = ['./img_list.pkl', './label_list.pkl']
train_dataset_path = DATASET_PATH + '/train/train_data'
if nsml.IS_ON_NSML:
# Caching file
nsml.cache(train_data_loader,
data_path=train_dataset_path,
output_path=output_path)
else:
train_dataset_path = config.debug_data
train_data_loader(train_dataset_path, output_path=output_path)
with open(output_path[0], 'rb') as img_f:
img_list = pickle.load(img_f)
with open(output_path[1], 'rb') as label_f:
label_list = pickle.load(label_f)
queries, references, queries_img, reference_img \
= convert_to_query_db_data_for_generator(img_list, label_list, input_shape, config.dev_querynum, config.dev_referencenum)
print("mAP devset : query(%d), reference(%d) " %
(len(queries), len(references)))
model = LadderModule()
if config.gpu:
model.cuda()
optimizer = optim.Adam(model.parameters(), lr=config.lr)
class_to_save = ClassToSave()
bind_model(model, class_to_save, optimizer=optimizer)
# test mode
if config.pause:
nsml.paused(scope=locals())
# training mode
if mode == 'train':
preprocessed_file = ['./processed.pt']
nsml.cache(preprocess,
output_path=preprocessed_file,
data=data_loader(DATASET_PATH))
dataset = torch.load(preprocessed_file[0])
training_data = DataLoader(dataset['train'],
batch_size,
shuffle=True,
num_workers=4)
testing_data = DataLoader(dataset['test'],
batch_size,
shuffle=False,
num_workers=4)
for epoch in range(config.epochs):
running_loss, running_acc = 0, 0
num_runs = 0
model.train()
