update postprocess_episodes for base_action, skip_mirrored_data option

这个提交包含在:
Zipeng Fu
2023-11-01 03:54:06 -04:00
父节点 7678af7021
当前提交 e8aa7976ba
共有 4 个文件被更改,包括 41 次插入 和 11 次删除

查看文件

@ -73,6 +73,7 @@ def get_args_parser():
parser.add_argument('--save_every', action='store', type=int, default=500, help='save_every', required=False) parser.add_argument('--save_every', action='store', type=int, default=500, help='save_every', required=False)
parser.add_argument('--resume_ckpt_path', action='store', type=str, help='load_ckpt_path', required=False) parser.add_argument('--resume_ckpt_path', action='store', type=str, help='load_ckpt_path', required=False)
parser.add_argument('--no_encoder', action='store_true') parser.add_argument('--no_encoder', action='store_true')
parser.add_argument('--skip_mirrored_data', action='store_true')
return parser return parser

查看文件

@ -127,7 +127,7 @@ def main(args):
print() print()
exit() exit()
train_dataloader, val_dataloader, stats, _ = load_data(dataset_dir, name_filter, camera_names, batch_size_train, batch_size_val, args['chunk_size']) train_dataloader, val_dataloader, stats, _ = load_data(dataset_dir, name_filter, camera_names, batch_size_train, batch_size_val, args['chunk_size'], args['skip_mirrored_data'])
# save dataset stats # save dataset stats
stats_path = os.path.join(ckpt_dir, f'dataset_stats.pkl') stats_path = os.path.join(ckpt_dir, f'dataset_stats.pkl')
@ -475,6 +475,7 @@ if __name__ == '__main__':
parser.add_argument('--validate_every', action='store', type=int, default=500, help='validate_every', required=False) parser.add_argument('--validate_every', action='store', type=int, default=500, help='validate_every', required=False)
parser.add_argument('--save_every', action='store', type=int, default=500, help='save_every', required=False) parser.add_argument('--save_every', action='store', type=int, default=500, help='save_every', required=False)
parser.add_argument('--resume_ckpt_path', action='store', type=str, help='resume_ckpt_path', required=False) parser.add_argument('--resume_ckpt_path', action='store', type=str, help='resume_ckpt_path', required=False)
parser.add_argument('--skip_mirrored_data', action='store_true')
# for ACT # for ACT
parser.add_argument('--kl_weight', action='store', type=int, help='KL Weight', required=False) parser.add_argument('--kl_weight', action='store', type=int, help='KL Weight', required=False)

查看文件

@ -15,8 +15,8 @@ e = IPython.embed
JOINT_NAMES = ["waist", "shoulder", "elbow", "forearm_roll", "wrist_angle", "wrist_rotate"] JOINT_NAMES = ["waist", "shoulder", "elbow", "forearm_roll", "wrist_angle", "wrist_rotate"]
STATE_NAMES = JOINT_NAMES + ["gripper"] STATE_NAMES = JOINT_NAMES + ["gripper"]
MIRROR_STATE_MULTIPLY = np.array([-1, 1, 1, -1, 1, -1, 1])[None, ...].astype('float32') MIRROR_STATE_MULTIPLY = np.array([-1, 1, 1, -1, 1, -1, 1]).astype('float32')
MIRROR_BASE_MULTIPLY = np.array([1, -1]).astype('float32')
def load_hdf5(dataset_dir, dataset_name): def load_hdf5(dataset_dir, dataset_name):
dataset_path = os.path.join(dataset_dir, dataset_name + '.hdf5') dataset_path = os.path.join(dataset_dir, dataset_name + '.hdf5')
@ -32,8 +32,13 @@ def load_hdf5(dataset_dir, dataset_name):
image_dict = dict() image_dict = dict()
for cam_name in root[f'/observations/images/'].keys(): for cam_name in root[f'/observations/images/'].keys():
image_dict[cam_name] = root[f'/observations/images/{cam_name}'][()] image_dict[cam_name] = root[f'/observations/images/{cam_name}'][()]
if 'base_action' in root.keys():
print('base_action exists')
base_action = root['/base_action'][()]
else:
base_action = None
return qpos, qvel, action, image_dict return qpos, qvel, action, base_action, image_dict
def main(args): def main(args):
dataset_dir = args['dataset_dir'] dataset_dir = args['dataset_dir']
@ -42,22 +47,40 @@ def main(args):
for episode_idx in range(num_episodes): for episode_idx in range(num_episodes):
dataset_name = f'episode_{episode_idx}' dataset_name = f'episode_{episode_idx}'
qpos, qvel, action, image_dict = load_hdf5(dataset_dir, dataset_name) qpos, qvel, action, base_action, image_dict = load_hdf5(dataset_dir, dataset_name)
# process proprioception # process proprioception
qpos = np.concatenate([qpos[:, 7:] * MIRROR_STATE_MULTIPLY, qpos[:, :7] * MIRROR_STATE_MULTIPLY], axis=1) qpos = np.concatenate([qpos[:, 7:] * MIRROR_STATE_MULTIPLY, qpos[:, :7] * MIRROR_STATE_MULTIPLY], axis=1)
qvel = np.concatenate([qvel[:, 7:] * MIRROR_STATE_MULTIPLY, qvel[:, :7] * MIRROR_STATE_MULTIPLY], axis=1) qvel = np.concatenate([qvel[:, 7:] * MIRROR_STATE_MULTIPLY, qvel[:, :7] * MIRROR_STATE_MULTIPLY], axis=1)
action = np.concatenate([action[:, 7:] * MIRROR_STATE_MULTIPLY, action[:, :7] * MIRROR_STATE_MULTIPLY], axis=1) action = np.concatenate([action[:, 7:] * MIRROR_STATE_MULTIPLY, action[:, :7] * MIRROR_STATE_MULTIPLY], axis=1)
if base_action is not None:
base_action = base_action * MIRROR_BASE_MULTIPLY
# mirror image obs # mirror image obs
if 'left_wrist' in image_dict.keys():
image_dict['left_wrist'], image_dict['right_wrist'] = image_dict['right_wrist'][:, :, ::-1], image_dict['left_wrist'][:, :, ::-1] image_dict['left_wrist'], image_dict['right_wrist'] = image_dict['right_wrist'][:, :, ::-1], image_dict['left_wrist'][:, :, ::-1]
elif 'cam_left_wrist':
image_dict['cam_left_wrist'], image_dict['cam_right_wrist'] = image_dict['cam_right_wrist'][:, :, ::-1], image_dict['cam_left_wrist'][:, :, ::-1]
else:
raise Exception('No left_wrist or cam_left_wrist in image_dict')
if 'top' in image_dict.keys():
image_dict['top'] = image_dict['top'][:, :, ::-1] image_dict['top'] = image_dict['top'][:, :, ::-1]
elif 'cam_high' in image_dict.keys():
image_dict['cam_high'] = image_dict['cam_high'][:, :, ::-1]
else:
raise Exception('No top or cam_high in image_dict')
# saving # saving
data_dict = { data_dict = {
'/observations/qpos': qpos, '/observations/qpos': qpos,
'/observations/qvel': qvel, '/observations/qvel': qvel,
'/action': action, '/action': action,
'/base_action': base_action,
} if base_action is not None else {
'/observations/qpos': qpos,
'/observations/qvel': qvel,
'/action': action,
} }
for cam_name in image_dict.keys(): for cam_name in image_dict.keys():
data_dict[f'/observations/images/{cam_name}'] = image_dict[cam_name] data_dict[f'/observations/images/{cam_name}'] = image_dict[cam_name]
@ -78,6 +101,8 @@ def main(args):
qpos = obs.create_dataset('qpos', (max_timesteps, 14)) qpos = obs.create_dataset('qpos', (max_timesteps, 14))
qvel = obs.create_dataset('qvel', (max_timesteps, 14)) qvel = obs.create_dataset('qvel', (max_timesteps, 14))
action = root.create_dataset('action', (max_timesteps, 14)) action = root.create_dataset('action', (max_timesteps, 14))
if base_action is not None:
base_action = root.create_dataset('base_action', (max_timesteps, 2))
for name, array in data_dict.items(): for name, array in data_dict.items():
root[name][...] = array root[name][...] = array

查看文件

@ -146,16 +146,19 @@ def get_norm_stats(dataset_path_list):
return stats, all_episode_len return stats, all_episode_len
def find_all_hdf5(dataset_dir): def find_all_hdf5(dataset_dir, skip_mirrored_data):
hdf5_files = [] hdf5_files = []
for root, dirs, files in os.walk(dataset_dir): for root, dirs, files in os.walk(dataset_dir):
for filename in fnmatch.filter(files, '*.hdf5'): for filename in fnmatch.filter(files, '*.hdf5'):
if skip_mirrored_data and 'mirror' in filename:
continue
hdf5_files.append(os.path.join(root, filename)) hdf5_files.append(os.path.join(root, filename))
print(f'Found {len(hdf5_files)} hdf5 files')
return hdf5_files return hdf5_files
def load_data(dataset_dir, name_filter, camera_names, batch_size_train, batch_size_val, chunk_size): def load_data(dataset_dir, name_filter, camera_names, batch_size_train, batch_size_val, chunk_size, skip_mirrored_data=False):
dataset_path_list = find_all_hdf5(dataset_dir) dataset_path_list = find_all_hdf5(dataset_dir, skip_mirrored_data)
dataset_path_list = [n for n in dataset_path_list if name_filter(n)] dataset_path_list = [n for n in dataset_path_list if name_filter(n)]
num_episodes = len(dataset_path_list) num_episodes = len(dataset_path_list)