print('Completed generating {} depth samples for sequence={}'.format(
n_generated, sequence_dirpath))
output_sparse_depth_filepath = \
TRAIN_SPARSE_DEPTH_OUTPUT_FILEPATH[:-4] + '-' + seq_id + '.txt'
output_validity_map_filepath = \
TRAIN_VALIDITY_MAP_OUTPUT_FILEPATH[:-4] + '-' + seq_id + '.txt'
output_semi_dense_depth_filepath = \
TRAIN_SEMI_DENSE_DEPTH_OUTPUT_FILEPATH[:-4] + '-' + seq_id + '.txt'
output_ground_truth_filepath = \
TRAIN_GROUND_TRUTH_OUTPUT_FILEPATH[:-4] + '-' + seq_id + '.txt'
print('Storing {} sparse depth file paths into: {}'.format(
len(output_sequence_sparse_depth_paths), output_sparse_depth_filepath))
data_utils.write_paths(output_sparse_depth_filepath,
output_sequence_sparse_depth_paths)
print('Storing {} validity map file paths into: {}'.format(
len(output_sequence_validity_map_paths), output_validity_map_filepath))
data_utils.write_paths(output_validity_map_filepath,
output_sequence_validity_map_paths)
print('Storing {} semi dense depth file paths into: {}'.format(
len(output_sequence_semi_dense_depth_paths),
output_semi_dense_depth_filepath))
data_utils.write_paths(output_semi_dense_depth_filepath,
output_sequence_semi_dense_depth_paths)
print('Storing {} ground truth file paths into: {}'.format(
len(output_sequence_ground_truth_paths), output_ground_truth_filepath))
data_utils.write_paths(output_ground_truth_filepath,
continue
else:
train_image_clean_output_paths.append(image_output_path)
train_sparse_depth_clean_output_paths.append(sparse_depth_output_path)
train_validity_map_clean_output_paths.append(validity_map_output_path)
train_ground_truth_clean_output_paths.append(ground_truth_output_path)
train_intrinsics_clean_output_paths.append(intrinsics_output_path)
sys.stdout.write(
'Processed {}/{} examples \r'.format(idx + 1, n_sample))
sys.stdout.flush()
print('Storing training image file paths into: %s' %
TRAIN_IMAGE_OUTPUT_FILEPATH)
data_utils.write_paths(
TRAIN_IMAGE_OUTPUT_FILEPATH, train_image_output_paths)
print('Storing training sparse depth file paths into: %s' %
TRAIN_SPARSE_DEPTH_OUTPUT_FILEPATH)
data_utils.write_paths(
TRAIN_SPARSE_DEPTH_OUTPUT_FILEPATH, train_sparse_depth_output_paths)
print('Storing training validity map file paths into: %s' %
TRAIN_VALIDITY_MAP_OUTPUT_FILEPATH)
data_utils.write_paths(
TRAIN_VALIDITY_MAP_OUTPUT_FILEPATH, train_validity_map_output_paths)
print('Storing training ground truth file paths into: %s' %
TRAIN_GROUND_TRUTH_OUTPUT_FILEPATH)
data_utils.write_paths(
TRAIN_GROUND_TRUTH_OUTPUT_FILEPATH, train_ground_truth_output_paths)
assert (len(stereo_flow_image0_paths) == len(stereo_flow_disparity_paths))
# Split KITTI 2012 dataset into training and test sets
stereo_flow_train_image0_paths = stereo_flow_image0_paths[0:160]
stereo_flow_train_image1_paths = stereo_flow_image1_paths[0:160]
stereo_flow_train_disparity_paths = stereo_flow_disparity_paths[0:160]
stereo_flow_test_image0_paths = stereo_flow_image0_paths[160:]
stereo_flow_test_image1_paths = stereo_flow_image1_paths[160:]
stereo_flow_test_disparity_paths = stereo_flow_disparity_paths[160:]
# Write all paths to disk
print(
'Storing all {} stereo flow left stereo images file paths into: {}'.format(
len(stereo_flow_image0_paths), STEREO_FLOW_ALL_IMAGE0_FILEPATH))
data_utils.write_paths(STEREO_FLOW_ALL_IMAGE0_FILEPATH,
stereo_flow_image0_paths)
print('Storing all {} stereo flow right stereo images file paths into: {}'.
format(len(stereo_flow_image1_paths), STEREO_FLOW_ALL_IMAGE1_FILEPATH))
data_utils.write_paths(STEREO_FLOW_ALL_IMAGE1_FILEPATH,
stereo_flow_image1_paths)
print('Storing all {} stereo flow ground truth disparity file paths into: {}'.
format(len(stereo_flow_disparity_paths),
STEREO_FLOW_ALL_DISPARITY_FILEPATH))
data_utils.write_paths(STEREO_FLOW_ALL_DISPARITY_FILEPATH,
stereo_flow_disparity_paths)
# Write training paths to disk
print('Storing {} training stereo flow left stereo images file paths into: {}'.
format(len(stereo_flow_train_image0_paths),
output_semi_dense_depth_path, \
output_dense_depth_path, \
output_ground_truth_path = result
# Collect filepaths
output_sparse_depth_paths.append(output_sparse_depth_path)
output_validity_map_paths.append(output_validity_map_path)
output_semi_dense_depth_paths.append(output_semi_dense_depth_path)
output_dense_depth_paths.append(output_dense_depth_path)
output_ground_truth_paths.append(output_ground_truth_path)
print('Completed generating {} depth samples for using KITTI sequence={} camera={}'.format(
n_vkitti_filepaths, kitti_sequence, camera_dirpath))
print('Storing sparse depth file paths into: %s' % OUTPUT_SPARSE_DEPTH_FILEPATH)
data_utils.write_paths(
OUTPUT_SPARSE_DEPTH_FILEPATH, output_sparse_depth_paths)
print('Storing validity map file paths into: %s' % OUTPUT_VALIDITY_MAP_FILEPATH)
data_utils.write_paths(
OUTPUT_VALIDITY_MAP_FILEPATH, output_validity_map_paths)
print('Storing semi dense depth file paths into: %s' % OUTPUT_SEMI_DENSE_DEPTH_FILEPATH)
data_utils.write_paths(
OUTPUT_SEMI_DENSE_DEPTH_FILEPATH, output_semi_dense_depth_paths)
print('Storing dense depth file paths into: %s' % OUTPUT_DENSE_DEPTH_FILEPATH)
data_utils.write_paths(
OUTPUT_DENSE_DEPTH_FILEPATH, output_dense_depth_paths)
print('Storing ground-truth depth file paths into: %s' % OUTPUT_GROUND_TRUTH_FILEPATH)
data_utils.write_paths(
print('Completed processing {} examples for sequence={}'.format(
len(pool_input), seq_dirpath))
print('Completed processing {} examples for density={}'.format(
n_sample, data_dirpath))
void_train_image_filepath, \
void_train_sparse_depth_filepath, \
void_train_validity_map_filepath, \
void_train_ground_truth_filepath, \
void_train_intrinsics_filepath = train_filepaths
print('Storing training image file paths into: %s' %
void_train_image_filepath)
data_utils.write_paths(void_train_image_filepath, train_image_outpaths)
print('Storing training sparse depth file paths into: %s' %
void_train_sparse_depth_filepath)
data_utils.write_paths(void_train_sparse_depth_filepath,
train_sparse_depth_outpaths)
print('Storing training validity map file paths into: %s' %
void_train_validity_map_filepath)
data_utils.write_paths(void_train_validity_map_filepath,
train_validity_map_outpaths)
print('Storing training groundtruth depth file paths into: %s' %
void_train_ground_truth_filepath)
data_utils.write_paths(void_train_ground_truth_filepath,
train_ground_truth_outpaths)
output_depth_dirpath = os.path.dirname(output_depth_path)
if not os.path.exists(output_depth_dirpath):
os.makedirs(output_depth_dirpath)
train_output_depth_paths.append(output_depth_path)
data_utils.save_depth(output_depth, output_depth_path)
n += 1
print('Processed {}/{} training examples \r'.format(n + 1, n_sample), end='\r')
except tf.errors.OutOfRangeError:
print('Currently processed {}/{} training examples'.format(n, n_sample))
break
print('Storing prediction for training file paths into: %s' % args.train_output_depth_path)
data_utils.write_paths(args.train_output_depth_path, train_output_depth_paths)
'''
Setup for validation and testing set
'''
# Load validation sparse depth and validity map paths from file
val_sparse_depth_paths = data_utils.read_paths(args.val_sparse_depth_path)
# Load testing sparse depth and validity map paths from file
test_sparse_depth_paths = data_utils.read_paths(args.test_sparse_depth_path)
val_output_depth_paths = []
test_output_depth_paths = []
modes = [
[