def main():
"""
need to specify save path: save_path_npy and save_path_txt
:return:
"""
args = make_parser().parse_args()
grasp_sampler_args = utils.read_checkpoint_args(args.grasp_sampler_folder)
grasp_sampler_args.is_train = False
grasp_evaluator_args = utils.read_checkpoint_args(
args.grasp_evaluator_folder)
grasp_evaluator_args.continue_train = True
estimator = grasp_estimator.GraspEstimator(grasp_sampler_args,
grasp_evaluator_args, args)
point_cloud = np.load(args.pc_path)
generated_grasps, generated_scores = estimator.generate_and_refine_grasps(
point_cloud)
grasp_pose_score = {
'grasp': np.array(generated_grasps),
'score': np.array(generated_scores)
}
np.save(args.npy_save_path, grasp_pose_score)
print('successfully saved %s' % args.npy_save_path)
with open(args.txt_save_path, 'w') as w:
for pose, score in zip(generated_grasps, generated_scores):
w.write(str(pose))
w.write(' ')
w.write(str(score) + '\n')
w.flush()
print('successfully saved %s' % args.txt_save_path)
def main(args):
parser = make_parser()
args = parser.parse_args()
grasp_sampler_args = utils.read_checkpoint_args(args.grasp_sampler_folder)
grasp_sampler_args.is_train = False
grasp_evaluator_args = utils.read_checkpoint_args(
args.grasp_evaluator_folder)
grasp_evaluator_args.continue_train = True
estimator = grasp_estimator.GraspEstimator(grasp_sampler_args,
grasp_evaluator_args, args)
if args.train_data:
grasp_sampler_args.dataset_root_folder = args.dataset_root_folder
grasp_sampler_args.num_grasps_per_object = 1
grasp_sampler_args.num_objects_per_batch = 1
dataset = DataLoader(grasp_sampler_args)
for i, data in enumerate(dataset):
generated_grasps, generated_scores = estimator.generate_and_refine_grasps(
data["pc"].squeeze())
return generated_grasps, generated_scores
else:
for npy_file in glob.glob(os.path.join(args.npy_folder, '*.npy'))[::-1]:
# Depending on your numpy version you may need to change allow_pickle
# from True to False.
data = np.load(npy_file, allow_pickle=True,
encoding="latin1").item()
depth = data['depth']
image = data['image']
K = data['intrinsics_matrix']
# Removing points that are farther than 1 meter or missing depth
# values.
#depth[depth == 0 or depth > 1] = np.nan
np.nan_to_num(depth, copy=False)
mask = np.where(np.logical_or(depth == 0, depth > 1))
depth[mask] = np.nan
pc, selection = backproject(depth,
K,
return_finite_depth=True,
return_selection=True)
pc_colors = image.copy()
pc_colors = np.reshape(pc_colors, [-1, 3])
pc_colors = pc_colors[selection, :]
# Smoothed pc comes from averaging the depth for 10 frames and removing
# the pixels with jittery depth between those 10 frames.
object_pc = data['smoothed_object_pc']
generated_grasps, generated_scores = estimator.generate_and_refine_grasps(
object_pc)
return generated_grasps, generated_scores
def configure(self, cfg: dict):
"""Additional class configuration
Parameters
----------
cfg : dict configuration dict, as sourced from the YAML file
"""
# Create a namespace from the config dict
# Since the GraspNet implementation uses a namespace
self.cfg_ns = SimpleNamespace(**self.cfg)
self.grasp_sampler_args = utils.read_checkpoint_args(
self.cfg_ns.grasp_sampler_folder)
self.grasp_sampler_args.is_train = False
self.grasp_evaluator_args = utils.read_checkpoint_args(
self.cfg_ns.grasp_evaluator_folder)
self.grasp_evaluator_args.continue_train = True
self.estimator = grasp_estimator.GraspEstimator(
self.grasp_sampler_args, self.grasp_evaluator_args, self.cfg_ns)
Returns:
grasp_success: list of binary numbers. 0 means that the grasp failed,
and 1 means that the grasp succeeded.
"""
raise NotImplementedError(
"The code for grasp evaluation is not released")
def __del__(self):
del self._grasp_reader
if __name__ == '__main__':
args = make_parser(sys.argv)
utils.mkdir(args.output_folder)
grasp_sampler_args = utils.read_checkpoint_args(args.grasp_sampler_folder)
grasp_sampler_args.is_train = False
grasp_evaluator_args = utils.read_checkpoint_args(
args.grasp_evaluator_folder)
grasp_evaluator_args.continue_train = True
if args.gradient_based_refinement:
args.num_refine_steps = 10
args.refinement = "gradient"
else:
args.num_refine_steps = 20
args.refinement = "sampling"
estimator = grasp_estimator.GraspEstimator(grasp_sampler_args,
grasp_evaluator_args, args)
evaluator = Evaluator(
cfg,