Error while generating synthetic data

[ceki]

Hi, I’m trying to generate dataset using offline pose generation example with custom USD file and getting this error. I hope you can help me:

Traceback (most recent call last): File "standalone_examples/replicator/offline_pose_generation/offline_doli_pose_generation.py", line 637, in <module> test=args.test, File "standalone_examples/replicator/offline_pose_generation/offline_doli_pose_generation.py", line 153, in __init__ self._setup_world() File "standalone_examples/replicator/offline_pose_generation/offline_doli_pose_generation.py", line 348, in _setup_world self.writer_helper.save_mesh_vertices(mesh_prim, coord_prim, prim_type, self._output_folder) File "/home/cenk/.local/share/ov/pkg/isaac_sim-2022.2.1/exts/omni.replicator.isaac/omni/replicator/isaac/scripts/writers/ycb_video_writer.py", line 184, in save_mesh_vertices points = get_mesh_vertices_relative_to(mesh_prim, coord_prim) File "/home/cenk/.local/share/ov/pkg/isaac_sim-2022.2.1/exts/omni.isaac.core/omni/isaac/core/utils/mesh.py", line 34, in get_mesh_vertices_relative_to vertices_tf_row_major = np.pad(vertices, ((0, 0), (0, 1)), constant_values=1.0) File "<__array_function__ internals>", line 6, in pad File "/home/cenk/.local/share/ov/pkg/isaac_sim-2022.2.1/kit/extscore/omni.kit.pip_archive/pip_prebundle/numpy/lib/arraypad.py", line 743, in pad pad_width = _as_pairs(pad_width, array.ndim, as_index=True) File "/home/cenk/.local/share/ov/pkg/isaac_sim-2022.2.1/kit/extscore/omni.kit.pip_archive/pip_prebundle/numpy/lib/arraypad.py", line 518, in _as_pairs return np.broadcast_to(x, (ndim, 2)).tolist() File "<__array_function__ internals>", line 6, in broadcast_to File "/home/cenk/.local/share/ov/pkg/isaac_sim-2022.2.1/kit/extscore/omni.kit.pip_archive/pip_prebundle/numpy/lib/stride_tricks.py", line 411, in broadcast_to return _broadcast_to(array, shape, subok=subok, readonly=True) File "/home/cenk/.local/share/ov/pkg/isaac_sim-2022.2.1/kit/extscore/omni.kit.pip_archive/pip_prebundle/numpy/lib/stride_tricks.py", line 350, in _broadcast_to op_flags=['readonly'], itershape=shape, order='C') ValueError: operands could not be broadcast together with remapped shapes [original->remapped]: (2,2) and requested shape (0,2) 2023-05-17 14:49:53 [13,962ms] [Warning] [carb.audio.context] 1 contexts were leaked ./python.sh: line 41: 23562 Segmentation fault (core dumped) $python_exe "$@" $args There was an error running python 

[ahaidu]

Hi there,

can you share the USD file that causes the issue? Did you check if other USD files are getting the same error, or only this particularone?

Best

[ceki]

I didn’t check with other USD files. I just imported a STL file into Isaac Sim and export as USD file.

[ceki]

dolly.usdc (1.4 MB)
dolly.usd (1.4 MB)

I’ve tried these two files. dolly.usdc exported from Blender and dolly.usd exported from Isaac.

[ahaidu]

Could you share the changes you did in the script and config file? Thanks!

[ceki]

The only change I made in offline_pose_generation.py:

 # Add the part to train the network on part_name = "dolly" ref_path = "/home/cenk/Desktop/dolly.usdc" prim_type = f"_{part_name[1:]}" path = "/World/" + prim_type mesh_path = path + "/" + prim_type name = "train_part" 

[ahaidu]

Can you check if you get an error if you set the DynamicObject as this:

train_part = DynamicObject( usd_path="/home/user/Downloads/dolly.usd", prim_path="/World/dolly", mesh_path="/World/dolly/_oli3/_oli3", name=name, position=np.array([0.0, 0.0, 0.0]), scale=np.array(config_data["OBJECT_SCALE"]), mass=1.0, ) 

The mesh_path needs to point to the mesh asset in the script:

[ceki]

Sorry for the late reply. I tried your way but it still didn’t work, I tried a lot but was able to do it just by editing like the example structure and naming the file _olly.usd.
For example, when I imported the 003_cracker_box model, the file was added as _03_cracker_box to the “World”, but when I imported the dolly.usd file, the file was added as dolly, not "_olly. Because of that i renamed the file to _olly.usd

Thanks for the help.

[ahaidu]

I am glad you managed to fix the issue. Yes, the script is currently tailored to the naming specifics of the YCB dataset meshes.