switch to not predict video, but training data

This commit is contained in:
Ruben van de Ven 2023-10-20 13:27:04 +02:00
parent f3ac903555
commit 2171dd459a

View file

@ -214,13 +214,13 @@ class PredictionServer:
prev_run_time = 0
while self.is_running.is_set():
timestep += 1
this_run_time = time.time()
logger.debug(f'test {prev_run_time - this_run_time}')
time.sleep(max(0, prev_run_time - this_run_time + .5))
prev_run_time = time.time()
# for timestep in range(init_timestep + 1, eval_scene.timesteps):
# this_run_time = time.time()
# logger.debug(f'test {prev_run_time - this_run_time}')
# time.sleep(max(0, prev_run_time - this_run_time + .5))
# prev_run_time = time.time()
# input_dict = eval_scene.get_clipped_input_dict(timestep, hyperparams['state'])
# TODO: see process_data.py on how to create a node, the provide nodes + incoming data columns
# data_columns = pd.MultiIndex.from_product([['position', 'velocity', 'acceleration'], ['x', 'y']])
# x = node_values[:, 0]
@ -239,62 +239,66 @@ class PredictionServer:
# node_data = pd.DataFrame(data_dict, columns=data_columns)
# node = Node(node_type=env.NodeType.PEDESTRIAN, node_id=node_id, data=node_data)
data = self.trajectory_socket.recv()
frame: Frame = pickle.loads(data)
trajectory_data = frame.trajectories # TODO: properly refractor
# trajectory_data = json.loads(data)
logger.debug(f"Receive {trajectory_data}")
if self.config.predict_training_data:
input_dict = eval_scene.get_clipped_input_dict(timestep, hyperparams['state'])
else:
data = self.trajectory_socket.recv()
frame: Frame = pickle.loads(data)
trajectory_data = frame.trajectories # TODO: properly refractor
# trajectory_data = json.loads(data)
logger.debug(f"Receive {trajectory_data}")
# class FakeNode:
# def __init__(self, node_type: NodeType):
# self.type = node_type
# class FakeNode:
# def __init__(self, node_type: NodeType):
# self.type = node_type
input_dict = {}
for identifier, trajectory in trajectory_data.items():
# if len(trajectory['history']) < 7:
# # TODO: these trajectories should still be in the output, but without predictions
# continue
input_dict = {}
for identifier, trajectory in trajectory_data.items():
# if len(trajectory['history']) < 7:
# # TODO: these trajectories should still be in the output, but without predictions
# continue
# TODO: modify this into a mapping function between JS data an the expected Node format
# node = FakeNode(online_env.NodeType.PEDESTRIAN)
history = [[h['x'], h['y']] for h in trajectory['history']]
history = np.array(history)
x = history[:, 0]
y = history[:, 1]
# TODO: calculate dt based on input
vx = derivative_of(x, 0.2) #eval_scene.dt
vy = derivative_of(y, 0.2)
ax = derivative_of(vx, 0.2)
ay = derivative_of(vy, 0.2)
# TODO: modify this into a mapping function between JS data an the expected Node format
# node = FakeNode(online_env.NodeType.PEDESTRIAN)
history = [[h['x'], h['y']] for h in trajectory['history']]
history = np.array(history)
x = history[:, 0]
y = history[:, 1]
# TODO: calculate dt based on input
vx = derivative_of(x, 0.2) #eval_scene.dt
vy = derivative_of(y, 0.2)
ax = derivative_of(vx, 0.2)
ay = derivative_of(vy, 0.2)
data_dict = {('position', 'x'): x[:],
('position', 'y'): y[:],
('velocity', 'x'): vx[:],
('velocity', 'y'): vy[:],
('acceleration', 'x'): ax[:],
('acceleration', 'y'): ay[:]}
data_columns = pd.MultiIndex.from_product([['position', 'velocity', 'acceleration'], ['x', 'y']])
data_dict = {('position', 'x'): x[:],
('position', 'y'): y[:],
('velocity', 'x'): vx[:],
('velocity', 'y'): vy[:],
('acceleration', 'x'): ax[:],
('acceleration', 'y'): ay[:]}
data_columns = pd.MultiIndex.from_product([['position', 'velocity', 'acceleration'], ['x', 'y']])
node_data = pd.DataFrame(data_dict, columns=data_columns)
node = Node(
node_type=online_env.NodeType.PEDESTRIAN,
node_id=identifier,
data=node_data,
first_timestep=timestep
)
node_data = pd.DataFrame(data_dict, columns=data_columns)
node = Node(
node_type=online_env.NodeType.PEDESTRIAN,
node_id=identifier,
data=node_data,
first_timestep=timestep
)
input_dict[node] = np.array([x[-1],y[-1],vx[-1],vy[-1],ax[-1],ay[-1]])
input_dict[node] = np.array([x[-1],y[-1],vx[-1],vy[-1],ax[-1],ay[-1]])
# print(input_dict)
# print(input_dict)
if not len(input_dict):
# skip if our input is empty
# TODO: we want to send out empty result...
if not len(input_dict):
# skip if our input is empty
# TODO: we want to send out empty result...
# And want to update the network
data = json.dumps({})
self.prediction_socket.send_string(data)
continue
data = json.dumps({})
self.prediction_socket.send_string(data)
continue
maps = None
if hyperparams['use_map_encoding']:
@ -311,12 +315,14 @@ class PredictionServer:
# robot_present_and_future += adjustment
start = time.time()
dists, preds = trajectron.incremental_forward(input_dict,
maps,
prediction_horizon=20, # TODO: make variable
num_samples=2, # TODO: make variable
robot_present_and_future=robot_present_and_future,
full_dist=True)
with warnings.catch_warnings():
warnings.simplefilter('ignore') # prevent deluge of UserWarning from torch's rrn.py
dists, preds = trajectron.incremental_forward(input_dict,
maps,
prediction_horizon=25, # TODO: make variable
num_samples=20, # TODO: make variable
robot_present_and_future=robot_present_and_future,
full_dist=True)
end = time.time()
logger.debug("took %.2f s (= %.2f Hz) w/ %d nodes and %d edges" % (end - start,
1. / (end - start), len(trajectron.nodes),
@ -365,7 +371,10 @@ class PredictionServer:
}
data = json.dumps(response)
logger.info(f"Total frame delay = {time.time()-frame.time}s ({len(trajectron.nodes)} nodes & {trajectron.scene_graph.get_num_edges()} edges. Trajectron: {end - start}s)")
if self.config.predict_training_data:
logger.info(f"Frame prediction: {len(trajectron.nodes)} nodes & {trajectron.scene_graph.get_num_edges()} edges. Trajectron: {end - start}s")
else:
logger.info(f"Total frame delay = {time.time()-frame.time}s ({len(trajectron.nodes)} nodes & {trajectron.scene_graph.get_num_edges()} edges. Trajectron: {end - start}s)")
self.prediction_socket.send_string(data)
logger.info('Stopping')