def main():
args = parse_args()
logging.info(args)
demo = BaseDemo(args)
if args.train:
demo.train()
if args.test:
demo.test()
if args.test_gt:
demo.test_gt()
def main():
args = parse_args()
logging.info(args)
demo = Demo(args)
if args.train:
demo.train_unsupervised()
if args.test:
demo.test_unsupervised()
if args.test_gt:
demo.test_gt_unsupervised()
def main():
logging.info('----------------------------------------------------------------')
logging.info('****************************************************************')
args = parse_args()
logging.info(args)
if args.data == 'kitti':
data = KittiData(args.data_path, args.train_proportion, args.test_proportion)
train_meta = data.train_meta
train_data = KittiDataLoader(train_meta, args.batch_size, args.image_heights,
args.image_widths, args.output_heights, args.output_widths,
args.num_scale, data_augment=False, shuffle=False)
test_meta = data.test_meta
test_data = KittiDataLoader(test_meta, args.batch_size, args.image_heights,
args.image_widths, args.output_heights, args.output_widths,
args.num_scale)
else:
print('Data not implemented yet')
return
if args.model == 'base':
model = BaseNet(args.image_channel, args.num_class)
elif args.model == 'base_3d':
model = Base3DNet(args.image_channel, args.depth_channel, args.num_class)
elif args.model == 'base_2stream':
model = Base2StreamNet(args.image_channel, args.depth_channel, args.num_class)
elif args.model == 'seg_3d':
model = Seg3DNet(args.image_channel, args.depth_channel, args.num_class)
else:
print('Model not implemented yet')
return
interface = DetectInterface(data, train_data, test_data, model, args.learning_rate,
args.train_epoch, args.test_interval, args.test_iteration,
args.save_interval, args.init_model_path, args.save_model_path,
args.tensorboard_path)
if args.train:
logging.info('Experiment: %s, training', args.exp_name)
interface.train()
elif args.test:
logging.info('Experiment: %s, testing all', args.exp_name)
interface.test_all()
elif args.visualize:
logging.info('Experiment: %s, visualizing', args.exp_name)
interface.visualize(args.image_name, args.depth_name, args.flow_name, args.box_name,
args.figure_path)
elif args.visualize_all:
logging.info('Experiment: %s, visualizing all', args.exp_name)
interface.visualize_all(args.image_list, args.figure_path)
else:
print('Unknown command')
return
def main():
logging.info(
'----------------------------------------------------------------')
logging.info(
'****************************************************************')
args = parse_args()
logging.info(args)
if args.data == 'vdrift':
data = VDriftData(args.data_path, args.batch_size, args.image_heights,
args.image_widths, args.output_height,
args.output_width, args.num_scale,
args.train_proportion, args.test_proportion,
args.show_statistics)
else:
print('Data not implemented yet')
return
if args.model == 'base':
model = BaseNet(args.image_channel, args.num_class)
elif args.model == 'base_3d':
model = Base3DNet(args.image_channel, args.depth_channel,
args.num_class)
elif args.model == 'base_2stream':
model = Base2StreamNet(args.image_channel, args.depth_channel,
args.num_class)
else:
print('Model not implemented yet')
return
interface = SegmentInterface(data, model, args.learning_rate,
args.train_iteration, args.test_iteration,
args.test_interval, args.save_interval,
args.init_model_path, args.save_model_path,
args.tensorboard_path)
if args.train:
logging.info('Experiment: %s, training', args.exp_name)
interface.train()
elif args.test:
logging.info('Experiment: %s, testing all', args.exp_name)
interface.test_all()
elif args.visualize:
logging.info('Experiment: %s, visualizing', args.exp_name)
interface.visualize(args.image_name, args.depth_name, args.flow_x_name,
args.flow_y_name, args.seg_name, args.figure_path)
elif args.visualize_all:
logging.info('Experiment: %s, visualizing all', args.exp_name)
interface.visualize_all(args.image_list, args.figure_path)
else:
print('Unknown command')
return
def main():
args = parse_args()
logging.info(args)
if args.data == 'mlt':
data = MLTData(args.batch_size, args.image_size, args.direction_type,
args.train_proportion, args.test_proportion,
args.show_statistics)
elif args.data == 'viper':
print('Not Implemented Yet')
return
data_test = DataTest(data)
data_test.test()
def main():
args = parse_args()
logging.info(args)
if args.data == 'kitti':
data = KittiData(args.data_path, args.train_proportion,
args.test_proportion)
train_meta = data.train_meta
train_data = KittiDataLoader(train_meta,
args.batch_size,
args.image_heights,
args.image_widths,
args.output_heights,
args.output_widths,
args.num_scale,
data_augment=True,
shuffle=True)
data_test = DataTest(train_data)
data_test.test()
test_meta = data.test_meta
test_data = KittiDataLoader(test_meta, args.batch_size,
args.image_heights, args.image_widths,
args.output_heights, args.output_widths,
args.num_scale)
data_test = DataTest(test_data)
data_test.test()
meta = {
'image': [args.image_name],
'depth': [args.depth_name],
'flow': [args.flow_name],
'box': [read_box(args.box_name)]
}
data = KittiDataLoader(meta, args.batch_size, args.image_heights,
args.image_widths, args.output_heights,
args.output_widths, args.num_scale)
data_test = DataTest(data)
data_test.test()
else:
print('Not Implemented Yet')
return
def main():
args = parse_args()
logging.info(args)
# args.data_path = '/media/yi/DATA/data-orig/kitti/training'
# args.image_name = '/media/yi/DATA/data-orig/kitti/training/image_2/007480.png'
# args.depth_name = '/media/yi/DATA/data-orig/kitti/training/disp_unsup/007480.png'
# args.flow_name = '/media/yi/DATA/data-orig/kitti/training/flow_unsup/007480.png'
# args.box_name = '/media/yi/DATA/data-orig/kitti/training/label_2/007480.txt'
if args.data == 'kitti':
data = KittiData(args.data_path, args.batch_size, args.image_heights, args.image_widths,
args.output_heights, args.output_widths, args.num_scale,
args.train_proportion, args.test_proportion, args.show_statistics)
else:
print('Not Implemented Yet')
return
data_test = DataTest(data)
data_test.test()
data_test.test_one_image(args.image_name, args.depth_name, args.flow_name, args.box_name)
def unit_test():
args = learning_args.parse_args()
logging.info(args)
data = MpiiData(args)
im = data.get_next_batch(data.train_images)
data.display(im)
def main():
logging.info('----------------------------------------------------------------')
logging.info('****************************************************************')
args = parse_args()
logging.info(args)
if args.data == 'mlt':
data = MLTData(args.batch_size, args.image_size, args.direction_type, args.train_proportion,
args.test_proportion, args.show_statistics)
elif args.data == 'viper':
print('Not Implemented Yet')
return
if args.model == 'base':
model = BaseNet(args.image_channel, args.num_class)
elif args.model == 'base_direct':
model = BaseDirectNet(args.image_channel, args.direction_dim, args.num_class)
elif args.model == 'base_3d':
model = Base3DNet(args.image_channel, args.depth_channel, args.num_class)
elif args.model == 'base_direct_3d':
model = BaseDirect3DNet(args.image_channel, args.depth_channel, args.direction_dim, args.num_class)
elif args.model == 'base_2stream':
model = Base2StreamNet(args.image_channel, args.depth_channel, args.num_class)
elif args.model == 'base_direct_2stream':
model = BaseDirect2StreamNet(args.image_channel, args.depth_channel, args.direction_dim, args.num_class)
elif args.model == 'hard_gt_attn':
model = HardGtAttnNet(args.image_size[0], args.image_channel, args.num_class)
elif args.model == 'hard_direct':
model = HardDirectNet(args.image_size[0], args.image_channel, args.direction_dim, args.num_class)
elif args.model == 'hard_gt_attn_3d':
model = HardGtAttn3DNet(args.image_size[0], args.image_channel, args.depth_channel, args.num_class)
elif args.model == 'hard_gt_attn_2stream':
model = HardGtAttn2StreamNet(args.image_size[0], args.image_channel, args.depth_channel, args.num_class)
elif args.model == 'soft_attn':
model = SoftAttnNet(args.attention_size, args.image_channel, args.num_class)
elif args.model == 'soft_direct':
model = SoftDirectNet(args.attention_size, args.image_channel, args.direction_dim, args.num_class)
elif args.model == 'soft_comb':
model = SoftCombNet(args.attention_size, args.image_channel, args.direction_dim, args.num_class)
elif args.model == 'soft_attn_3d':
model = SoftAttn3DNet(args.attention_size, args.image_channel, args.depth_channel, args.num_class)
elif args.model == 'soft_direct_3d':
model = SoftDirect3DNet(args.attention_size, args.image_channel, args.depth_channel, args.direction_dim, args.num_class)
elif args.model == 'soft_comb_3d':
model = SoftComb3DNet(args.attention_size, args.image_channel, args.depth_channel, args.direction_dim, args.num_class)
elif args.model == 'soft_attn_2stream':
model = SoftAttn2StreamNet(args.attention_size, args.image_channel, args.depth_channel, args.num_class)
elif args.model == 'soft_direct_2stream':
model = SoftDirect2StreamNet(args.attention_size, args.image_channel, args.depth_channel, args.direction_dim, args.num_class)
elif args.model == 'soft_comb_2stream':
model = SoftComb2StreamNet(args.attention_size, args.image_channel, args.depth_channel, args.direction_dim, args.num_class)
if args.attention_type == 'soft':
interface = SoftAttnInterface(data, model, args.learning_rate, args.train_iteration, args.test_iteration,
args.test_interval, args.save_interval, args.init_model_path,
args.save_model_path, args.tensorboard_path)
elif args.attention_type == 'hard':
interface = HardAttnInterface(data, model, args.learning_rate, args.train_iteration, args.test_iteration,
args.test_interval, args.save_interval, args.init_model_path,
args.save_model_path, args.tensorboard_path)
if args.train:
logging.info('Experiment: %s, training', args.exp_name)
interface.train()
elif args.test:
logging.info('Experiment: %s, testing all', args.exp_name)
interface.test_all()
elif args.visualize:
logging.info('Experiment: %s, visualizing', args.exp_name)
interface.visualize(args.image_name, args.depth_name, args.box_name, args.figure_path)
elif args.visualize_all:
logging.info('Experiment: %s, visualizing all', args.exp_name)
interface.visualize_all(args.image_list, args.figure_path)
def unit_test():
args = learning_args.parse_args()
logging.info(args)
data = BoxDataComplex(args)
im, motion, motion_r, motion_label, motion_label_r, seg_layer = data.get_next_batch()
data.display(im, motion, motion_r, seg_layer)
def unit_test():
args = learning_args.parse_args()
logging.info(args)
data = Kitti128Sample(args)
im = data.get_next_batch(data.test_images)
data.display(im)
def unit_test():
args = learning_args.parse_args()
logging.info(args)
data = ChairData(args)
im, motion, motion_label, seg_layer = data.get_next_batch(data.train_images)
data.display(im, motion, seg_layer)
def main():
args = parse_args()
logging.info(args)
demo = BaseDemo(args)
demo.compare()
def unit_test():
args = learning_args.parse_args()
logging.info(args)
data = Robot64Data(args)
im = data.get_next_batch(data.train_meta)
data.display(im)
def unit_test():
args = learning_args.parse_args()
logging.info(args)
data = MnistDataBidirect(args)
im, motion, motion_r, motion_label, motion_label_r, seg_layer = data.get_next_batch(data.train_images)
data.display(im, motion, motion_r, seg_layer)
