def setup_path(self, flags):
root_folder = flags.data_root
train_data = [
'art_painting_train_features.hdf5', 'cartoon_train_features.hdf5',
'photo_train_features.hdf5', 'sketch_train_features.hdf5'
]
val_data = [
'art_painting_val_features.hdf5', 'cartoon_val_features.hdf5',
'photo_val_features.hdf5', 'sketch_val_features.hdf5'
]
test_data = [
'art_painting_features.hdf5', 'cartoon_features.hdf5',
'photo_features.hdf5', 'sketch_features.hdf5'
]
self.train_paths = []
for data in train_data:
path = os.path.join(root_folder, data)
self.train_paths.append(path)
self.val_paths = []
for data in val_data:
path = os.path.join(root_folder, data)
self.val_paths.append(path)
unseen_index = flags.unseen_index
self.unseen_data_path = os.path.join(root_folder,
test_data[unseen_index])
self.train_paths.remove(self.train_paths[unseen_index])
self.val_paths.remove(self.val_paths[unseen_index])
if not os.path.exists(flags.logs):
os.mkdir(flags.logs)
flags_log = os.path.join(flags.logs, 'path_log.txt')
write_log(str(self.train_paths), flags_log)
write_log(str(self.val_paths), flags_log)
write_log(str(self.unseen_data_path), flags_log)
self.batImageGenTrains = []
for train_path in self.train_paths:
batImageGenTrain = BatchImageGenerator(flags=flags,
file_path=train_path,
stage='train',
b_unfold_label=False)
self.batImageGenTrains.append(batImageGenTrain)
self.batImageGenVals = []
for val_path in self.val_paths:
batImageGenVal = BatchImageGenerator(flags=flags,
file_path=val_path,
stage='val',
b_unfold_label=True)
self.batImageGenVals.append(batImageGenVal)
def heldout_test(self, flags):
# load the best model in the validation data
model_path = os.path.join(flags.model_path, 'best_model.tar')
self.load_state_dict(state_dict=model_path)
# test
batImageGenTest = BatchImageGenerator(flags=flags,
file_path=self.unseen_data_path,
stage='test',
b_unfold_label=False)
test_images = batImageGenTest.images
threshold = 100
n_slices_test = len(test_images) / threshold
indices_test = []
for per_slice in range(n_slices_test - 1):
indices_test.append(
len(test_images) * (per_slice + 1) / n_slices_test)
test_image_splits = np.split(test_images,
indices_or_sections=indices_test)
# Verify the splits are correct
test_image_splits_2_whole = np.concatenate(test_image_splits)
assert np.all(test_images == test_image_splits_2_whole)
# split the test data into splits and test them one by one
predictions = []
self.network.eval()
for test_image_split in test_image_splits:
images_test = Variable(
torch.from_numpy(np.array(test_image_split,
dtype=np.float32))).cuda()
outputs, end_points = self.network(images_test)
pred = end_points['Predictions']
pred = pred.cpu().data.numpy()
predictions.append(pred)
# concatenate the test predictions first
predictions = np.concatenate(predictions)
# accuracy
accuracy = accuracy_score(y_true=batImageGenTest.labels,
y_pred=np.argmax(predictions, -1))
flags_log = os.path.join(flags.logs, 'heldout_test_log.txt')
write_log(accuracy, flags_log)
def test(self,
flags,
ite,
log_prefix,
log_dir='logs/',
batImageGenTest=None):
# switch on the network test mode
self.network.eval()
if batImageGenTest is None:
batImageGenTest = BatchImageGenerator(flags=flags,
file_path='',
stage='test',
b_unfold_label=True)
images_test = batImageGenTest.images
labels_test = batImageGenTest.labels
threshold = 50
if len(images_test) > threshold:
n_slices_test = len(images_test) / threshold
indices_test = []
for per_slice in range(n_slices_test - 1):
indices_test.append(
len(images_test) * (per_slice + 1) / n_slices_test)
test_image_splits = np.split(images_test,
indices_or_sections=indices_test)
# Verify the splits are correct
test_image_splits_2_whole = np.concatenate(test_image_splits)
assert np.all(images_test == test_image_splits_2_whole)
# split the test data into splits and test them one by one
test_image_preds = []
for test_image_split in test_image_splits:
images_test = Variable(
torch.from_numpy(
np.array(test_image_split, dtype=np.float32))).cuda()
outputs, end_points = self.network(images_test)
predictions = end_points['Predictions']
predictions = predictions.cpu().data.numpy()
test_image_preds.append(predictions)
# concatenate the test predictions first
predictions = np.concatenate(test_image_preds)
else:
images_test = Variable(
torch.from_numpy(np.array(images_test,
dtype=np.float32))).cuda()
outputs, end_points = self.network(images_test)
predictions = end_points['Predictions']
predictions = predictions.cpu().data.numpy()
accuracy = compute_accuracy(predictions=predictions,
labels=labels_test)
print('----------accuracy test----------:', accuracy)
if not os.path.exists(log_dir):
os.mkdir(log_dir)
log_path = os.path.join(log_dir, '{}.txt'.format(log_prefix))
write_log(str('ite:{}, accuracy:{}'.format(ite, accuracy)),
log_path=log_path)
# switch on the network train mode after test
self.network.train()
return accuracy
def setup_path(self, flags):
root_folder = flags.data_root #provide this as input: root_folder = '/Users/pulkit/Desktop/PACS/'
data_text_files = ['/Users/pulkit/Desktop/data_paths/art_painting.txt',
'/Users/pulkit/Desktop/data_paths/cartoon.txt',
'/Users/pulkit/Desktop/data_paths/photo.txt',
'/Users/pulkit/Desktop/data_paths/sketch.txt']
self.train_paths = ['/Users/pulkit/Desktop/data_paths/art_painting.txt',
'/Users/pulkit/Desktop/data_paths/cartoon.txt',
'/Users/pulkit/Desktop/data_paths/photo.txt',
'/Users/pulkit/Desktop/data_paths/sketch.txt']
self.val_paths = ['/Users/pulkit/Desktop/data_paths/art_painting.txt',
'/Users/pulkit/Desktop/data_paths/cartoon.txt',
'/Users/pulkit/Desktop/data_paths/photo.txt',
'/Users/pulkit/Desktop/data_paths/sketch.txt']
unseen_index = flags.unseen_index
self.train_paths.remove(self.train_paths[unseen_index])
self.val_paths.remove(self.val_paths[unseen_index])
self.unseen_data_path = data_text_files[unseen_index]
if not os.path.exists(flags.logs):
os.makedirs(flags.logs)
flags_log = os.path.join(flags.logs, 'path_log.txt')
write_log(str(self.train_paths), flags_log)
write_log(str(self.val_paths), flags_log)
write_log(str(self.unseen_data_path), flags_log)
# self.batImageGenTrains = []
# self.batImageGenVals = []
# self.batImageGenTest = []
# Get all the data at once in the dataset from the DataSet class
# and then sample data at every iteration of training
for path in data_text_files:
dataset = BatchImageGenerator(root=root_folder, flist=path,
transform=transforms.Compose([transforms.Resize(227),
#transforms.RandomResizedCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])]))
dataset_size = len(dataset)
indices = list(range(dataset_size))
until_train = int(np.floor(0.8 * dataset_size))
until_val = int(np.floor(0.9 * dataset_size))
np.random.shuffle(indices)
train_indices, val_indices, test_indices = indices[:until_train], indices[until_train:until_val],\
indices[until_val:]
self.TrainMetaData.append([dataset, train_indices])
self.TestMetaData.append([dataset, val_indices])
self.ValidMetaData.append([dataset, test_indices])