def __init__(self, args):
# initialize environment
self.args = args
if args.vae_dist_help:
load_vaes(args)
self.env = make_env(args)
self.args.timesteps = self.env.env.env.spec.max_episode_steps
self.env_test = make_env(args)
self.info = []
self.test_rollouts = 12
# get current policy from path (restore tf session + graph)
self.play_dir = args.play_path
self.play_epoch = args.play_epoch
self.meta_path = "{}saved_policy-{}.meta".format(self.play_dir, self.play_epoch)
self.sess = tf.Session()
self.saver = tf.train.import_meta_graph(self.meta_path)
self.saver.restore(self.sess, tf.train.latest_checkpoint(self.play_dir))
graph = tf.get_default_graph()
self.raw_obs_ph = graph.get_tensor_by_name("raw_obs_ph:0")
self.pi = graph.get_tensor_by_name("main/policy/net/pi/Tanh:0")
def flat_entries(bboxes_list, ppair):
return np.concatenate([bboxes_list.ravel(), ppair.ravel()])
if __name__ == "__main__":
args = get_args()
# create data folder if it does not exist, corresponding folders, and files where to store data
this_file_dir = os.path.dirname(os.path.abspath(__file__)) + '/'
base_data_dir = this_file_dir + 'data/'
env_data_dir = base_data_dir + args.env + '/'
make_dir(env_data_dir, clear=False)
if args.vae_dist_help:
load_vaes(args)
load_field_parameters(args)
env = make_temp_env(args)
field_names = ['ppair', 'bbox', 'distance']
csv_file_path = env_data_dir + 'distances.csv'
csv_file_path_val = env_data_dir + 'distances_val.csv'
csv_file_path_test = env_data_dir + 'distances_test.csv'
for csv_path in [csv_file_path, csv_file_path_val, csv_file_path_test]:
if os.path.exists(csv_path):
os.remove(csv_path)
with open(csv_path, 'w') as csv_file:
writer = csv.DictWriter(csv_file, fieldnames=field_names)
writer.writeheader()
#create about 20 000 samples for each step
z_pres, z_depth, z_scale, z_pos = z_pres.detach().cpu().numpy(), z_depth.detach().cpu().numpy(), \
z_scale.detach().cpu().numpy(), z_pos.detach().cpu().numpy()
return z_pres, z_depth, z_scale, z_pos
if __name__ == "__main__":
args = get_args()
# create data folder if it does not exist, corresponding folders, and files where to store data
this_file_dir = os.path.dirname(os.path.abspath(__file__)) + '/'
base_data_dir = this_file_dir + 'data/'
env_data_dir = base_data_dir + args.env + '/'
make_dir(env_data_dir, clear=False)
assert args.vae_dist_help
load_vaes(args, doing_inference=True)
load_field_parameters(args)
env = gym.make(args.env)
pres = []
scale = []
pos = []
for rs in range(10):
env.reset()
image = take_image_objects(None, args.img_size, direct_env=env.env)
#im_current = Image.fromarray(image.astype(np.uint8))
#im_current.save('env_image_for_vae.png')
z_pres, z_depth, z_scale, z_pos = extract_info(np.array([image]), args)
pres.append(z_pres[0])
scale.append(z_scale[0])