def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"--input-type",
default="blind",
choices=["blind", "rgb", "depth", "rgbd"],
)
parser.add_argument("--evaluation", type=str, required=True, choices=["local", "remote"])
config_paths = os.environ["CHALLENGE_CONFIG_FILE"]
parser.add_argument("--model-path", default="", type=str)
args = parser.parse_args()
config = get_config('configs/ddppo_pointnav.yaml',
['BASE_TASK_CONFIG_PATH', config_paths]).clone()
config.defrost()
config.TORCH_GPU_ID = 0
config.INPUT_TYPE = args.input_type
config.MODEL_PATH = args.model_path
config.RANDOM_SEED = 7
config.freeze()
agent = DDPPOAgent(config)
if args.evaluation == "local":
challenge = habitat.Challenge(eval_remote=False)
else:
challenge = habitat.Challenge(eval_remote=True)
challenge.submit(agent)
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--evaluation", type=str, required=True, choices=["local", "remote"])
args = parser.parse_args()
config_paths = os.environ["CHALLENGE_CONFIG_FILE"]
config = habitat.get_config(config_paths)
agent = RandomAgent(task_config=config)
if args.evaluation == "local":
challenge = habitat.Challenge(eval_remote=False)
else:
challenge = habitat.Challenge(eval_remote=True)
challenge.submit(agent)
def run_cfg(cfg, uuid):
if cfg['eval_kwargs']['exp_path'] is not None:
# Process exp path
exp_paths = [cfg['eval_kwargs']['exp_path']]
# Set up config with the first exp only
metadata_dir = get_subdir(exp_paths[0], 'metadata')
config_path = os.path.join(metadata_dir, 'config.json')
# Load config
with open(config_path) as config:
config_data = json.load(config)
# Update configs
config_data['uuid'] += '_benchmark' + uuid
config_data['cfg']['saving']['log_dir'] += '/benchmark'
config_data['cfg']['saving']['visdom_log_file'] = os.path.join(config_data['cfg']['saving']['log_dir'], 'visdom_logs.json')
config_data['cfg']['learner']['test'] = True
if cfg['eval_kwargs']['overwrite_configs']:
config_data['cfg'] = update_dict_deepcopy(config_data['cfg'], cfg)
set_seed(config_data['cfg']['training']['seed'])
# Get checkpoints
ckpt_paths = []
for exp_path in exp_paths:
ckpts_dir = get_subdir(exp_path, 'checkpoints')
ckpt_path = os.path.join(ckpts_dir, 'ckpt-latest.dat')
ckpt_paths.append(ckpt_path)
else:
config_data = { 'cfg': cfg, 'uuid': uuid }
ckpt_paths = [None]
exp_paths = [LOG_DIR]
if 'eval_kwargs' in cfg and 'debug' in cfg['eval_kwargs']:
if cfg['eval_kwargs']['debug']:
config_data['cfg']['saving']['logging_type'] = 'visdom'
config_data['cfg']['saving']['save_eval_videos'] = True
else:
config_data['cfg']['saving']['save_eval_videos'] = False
print(pprint.pformat(config_data))
print('Loaded:', config_data['uuid'])
agent = HabitatAgent(ckpt_path=ckpt_paths[0], config_data=config_data)
if cfg['eval_kwargs']['challenge']:
challenge = habitat.Challenge()
challenge.submit(agent)
else:
benchmark = habitat.Benchmark(config_file=cfg['eval_kwargs']['benchmark_config'], config_dir='/')
metrics = benchmark.evaluate(agent, cfg['eval_kwargs']['benchmark_episodes'])
agent.finish_benchmark(metrics)
benchmark._env.close()
everything = update_dict_deepcopy(metrics, config_data)
patience, unstuck_dist = config_data['cfg']['learner']['backout']['patience'], config_data['cfg']['learner']['backout']['unstuck_dist']
write_location = os.path.join(exp_paths[0], f'benchmark_data_p{patience}_d{unstuck_dist}.json')
with open(write_location, 'w') as outfile:
json.dump(everything, outfile)
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--agent-class", type=str, default="GoalFollower")
args = parser.parse_args()
agent = get_agent_cls(args.agent_class)(success_distance=0.2)
challenge = habitat.Challenge()
challenge.submit(agent)
def main():
# ! Note, there's some additional config not ported from dev setup, but those choices shouldn't matter...
parser = argparse.ArgumentParser()
parser.add_argument("--evaluation",
type=str,
required=True,
choices=["local", "remote"])
config_paths = os.environ["CHALLENGE_CONFIG_FILE"]
parser.add_argument("--model-path", default="", type=str)
parser.add_argument("--config-path",
type=str,
required=True,
default="configs/aux_objectnav.yaml")
args = parser.parse_args()
DEFAULT_CONFIG = "configs/il_objectnav.yaml"
config = get_config([DEFAULT_CONFIG, args.config_path],
['BASE_TASK_CONFIG_PATH', config_paths]).clone()
config.defrost()
config.TORCH_GPU_ID = 0
config.MODEL_PATH = args.model_path
seed = 7
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.random.manual_seed(seed)
config.RANDOM_SEED = 7
config.freeze()
torch.backends.cudnn.benchmark = False
agent = ILAgent(config)
if args.evaluation == "local":
challenge = habitat.Challenge(eval_remote=False)
else:
challenge = habitat.Challenge(eval_remote=True)
challenge.submit(agent)
def run_cfg(cfg, uuid):
if cfg['eval_kwargs']['exp_path'] is not None:
# Process exp path
exp_paths = [cfg['eval_kwargs']['exp_path']]
# Set up config with the first exp only
metadata_dir = get_subdir(exp_paths[0], 'metadata')
config_path = os.path.join(metadata_dir, 'config.json')
# Load config
with open(config_path) as config:
config_data = json.load(config)
# Update configs
config_data['uuid'] += '_benchmark' + uuid
config_data['cfg']['saving']['log_dir'] += '/benchmark'
config_data['cfg']['saving']['visdom_log_file'] = os.path.join(config_data['cfg']['saving']['log_dir'], 'visdom_logs.json')
config_data['cfg']['learner']['test'] = True
if cfg['eval_kwargs']['overwrite_configs']:
config_data['cfg'] = update_dict_deepcopy(config_data['cfg'], cfg)
set_seed(config_data['cfg']['training']['seed'])
# Get checkpoints
ckpt_paths = []
for exp_path in exp_paths:
ckpts_dir = get_subdir(exp_path, 'checkpoints')
ckpt_path = os.path.join(ckpts_dir, 'ckpt-latest.dat')
ckpt_paths.append(ckpt_path)
else:
config_data = { 'cfg': cfg, 'uuid': uuid }
ckpt_paths = [None]
exp_paths = [LOG_DIR]
if 'eval_kwargs' in cfg and 'debug' in cfg['eval_kwargs']:
if cfg['eval_kwargs']['debug']:
config_data['cfg']['saving']['logging_type'] = 'visdom'
config_data['cfg']['saving']['save_eval_videos'] = True
else:
config_data['cfg']['saving']['save_eval_videos'] = False
print(pprint.pformat(config_data))
print('Loaded:', config_data['uuid'])
agent = HabitatAgent(ckpt_path=ckpt_paths[0], config_data=config_data)
if cfg['eval_kwargs']['challenge']:
challenge = habitat.Challenge()
challenge.submit(agent)
else:
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"--input-type",
default="blind",
choices=["blind", "rgb", "depth", "rgbd"],
)
parser.add_argument("--model-path", default="", type=str)
args = parser.parse_args()
config = get_defaut_config()
config.INPUT_TYPE = args.input_type
config.MODEL_PATH = args.model_path
agent = PPOAgent(config)
challenge = habitat.Challenge()
challenge.submit(agent)
def get_action_space(self, request, context):
challenge = habitat.Challenge()
agent = unpack_for_grpc(request.SerializedEntity)
challenge.submit(agent)
return evaluation_pb2.Package(SerializedEntity=agent)
def main():
config_paths = os.environ["CHALLENGE_CONFIG_FILE"]
config = habitat.get_config(config_paths)
agent = RandomAgent(task_config=config)
challenge = habitat.Challenge()
challenge.submit(agent)
def main():
params = parse_args(default_files=('./habitat_submission.conf', ))
is_submission = (params.habitat_eval not in ['localtest', 'gendata'])
if params.seed > 0:
np.random.seed(params.seed)
random.seed(params.seed)
# Update for 2021 change in environment variable naming
if os.environ["TRACK_CONFIG_FILE"]:
print("Overwriting CHALLENGE_CONFIG_FILE with %s" %
os.environ["TRACK_CONFIG_FILE"])
os.environ["CHALLENGE_CONFIG_FILE"] = os.environ["TRACK_CONFIG_FILE"]
if not is_submission:
if params.habitat_config_file != "":
os.environ["CHALLENGE_CONFIG_FILE"] = params.habitat_config_file
elif params.overwrite_agent_noise >= 0.:
assert params.overwrite_agent_noise == 0. # need a separate file otherwise
print("Overwriting agent noise %f" % params.overwrite_agent_noise)
os.environ[
"CHALLENGE_CONFIG_FILE"] = './configs/challenge_pointnav_supervised_nonoise.yaml'
else:
os.environ[
"CHALLENGE_CONFIG_FILE"] = './configs/challenge_pointnav_supervised.yaml'
# parser = argparse.ArgumentParser()
# parser.add_argument("--evaluation", type=str, required=True, choices=["local", "remote"])
# parser.add_argument("--config", type=str, default='', required=False)
# args = parser.parse_args()
#
# if args.config != '':
# os.environ["CHALLENGE_CONFIG_FILE"] = args.config
config_paths = os.environ["CHALLENGE_CONFIG_FILE"]
config = habitat.get_config(config_paths)
if config.SIMULATOR.TYPE.startswith('Spot'):
# This will register the custom sim class
from spot import spotsim
# Update compressed episode json
jsonfile = config.DATASET.DATA_PATH
assert jsonfile.endswith('.gz')
jsonfile = jsonfile[:-3]
import subprocess
subprocess.call(['gzip -kf ' + jsonfile], shell=True)
# import ipdb as pdb
# pdb.set_trace()
# from habitat.sims.pyrobot import pyrobot
print("Using config file(s): %s" % (str(config_paths)))
logdir = params.logpath if params.logpath != '' \
else './temp/evals/{}-{}'.format(time.strftime('%m-%d-%H-%M-%S', time.localtime()), params.name) # this is only used for videos now, should pass it in to challenge eval
os.makedirs(logdir, exist_ok=True)
# agent = RandomAgent(task_config=config)
if params.habitat_eval in ["localtest", "gendata"]:
# # TODO Try to add additional noiseless sensors. Changes here have no effect!! Modify config file instead.
# config.defrost()
# # config.SIMULATOR.AGENT_0.SENSORS.append("NOISELESS_RGB_SENSOR")
# # config.SIMULATOR.NOISELESS_RGB_SENSOR = habitat.Config(config.SIMULATOR.RGB_SENSOR)
# # config.SIMULATOR.NOISELESS_RGB_SENSOR.NOISE_MODEL = ""
# # config.SIMULATOR.NOISELESS_RGB_SENSOR.NOISE_MODEL_KWARGS = habitat.Config()
# # config.SIMULATOR.AGENT_0.SENSORS.append("NOISELESS_DEPTH_SENSOR")
# # config.SIMULATOR.NOISELESS_DEPTH_SENSOR = habitat.Config(config.SIMULATOR.DEPTH_SENSOR)
# # config.SIMULATOR.NOISELESS_DEPTH_SENSOR.NOISE_MODEL = ""
# # config.SIMULATOR.NOISELESS_DEPTH_SENSOR.NOISE_MODEL_KWARGS = habitat.Config()
# config.SIMULATOR.AGENT_0.SENSORS.append("NOISELESS_RGB_SENSOR")
# config.SIMULATOR.NOISELESS_RGB_SENSOR = habitat.Config(config.SIMULATOR.RGB_SENSOR)
# config.SIMULATOR.NOISELESS_RGB_SENSOR.TYPE = "SecondHabitatSimRGBSensor"
# config.SIMULATOR.NOISELESS_RGB_SENSOR.NOISE_MODEL = ""
# config.SIMULATOR.NOISELESS_RGB_SENSOR.NOISE_MODEL_KWARGS = habitat.Config()
# config.freeze()
challenge = habitat.Challenge(eval_remote=False)
env = challenge._env
grid_cell_size = 0.05 # 5cm
map_size = (maps.COORDINATE_MAX - maps.COORDINATE_MIN) / grid_cell_size
assert config.TASK.TOP_DOWN_MAP.MAP_RESOLUTION == int(map_size)
if params.seed > 0:
env._sim.seed(params.seed)
env.seed(params.seed)
# agent = ShortestPathAgent(task_config=config, env=env)
if params.habitat_eval == "gendata":
assert len(params.data_map_sizes) > 0
# Initialize writers
data_filenames = [
os.path.join(
logdir, "agentplandata.m%d.%d-%d.tfrecords" %
(map_size, params.skip_first_n, params.num_episodes))
for map_size in params.data_map_sizes
]
# with statement for variable number of items
from contextlib import ExitStack
with ExitStack() as stack:
tfwriters = [
stack.enter_context(
tf.python_io.TFRecordWriter(data_filename))
for data_filename in data_filenames
]
agent = DSLAMAgent(task_config=config,
params=params,
env=env,
logdir=logdir,
tfwriters=tfwriters)
challenge.submit(agent,
num_episodes=params.num_episodes,
skip_first_n=params.skip_first_n)
else:
agent = DSLAMAgent(task_config=config,
params=params,
env=env,
logdir=logdir)
challenge.submit(agent,
num_episodes=params.num_episodes,
skip_first_n=params.skip_first_n)
elif params.habitat_eval == "local":
challenge = habitat.Challenge(eval_remote=False)
# if params.seed > 0:
# challenge._env._sim.seed(params.seed)
# challenge._env.seed(params.seed)
agent = DSLAMAgent(task_config=config,
params=params,
env=None,
logdir=logdir)
# agent = RandomAgent(task_config=config, params=params)
challenge.submit(agent) # , num_episodes=params.num_episodes)
else:
challenge = habitat.Challenge(eval_remote=True)
agent = DSLAMAgent(task_config=config,
params=params,
env=None,
logdir=logdir)
# agent = RandomAgent(task_config=config, params=params)
challenge.submit(agent)
def main():
agent = ForwardOnlyAgent()
challenge = habitat.Challenge()
challenge.submit(agent)
