A simple module for convenient usage of both NumPy and PyTorch point clouds. Functionality includes:
- Visualization in Plotly, Open3D, and Rerun with one function call.
- Tracks and enforces consistency between XYZ and (optional) RGB PyTorch data type and device.
- Methods for (un)normalization of RGB values.
To install arm_clouds, clone the repository and install with pip install -e ..
This repo includes classes for both single point clouds (PointCloud) and point cloud lists (PointCloudList).
An example PointCloud:
# Your XYZ and RGB data
xyz = torch.rand(3, 100)
rgb = torch.rand(3, 100) # optional
pc = PointCloud(xyz, rgb)
# One-call-visualization. Can also visualize in rerun and open3d.
pc.visualize_plotly()You can initialize a PointCloudList from a NumPy array or PyTorch tensor of all point clouds:
xyz = np.random.rand(100, 3, 25)
pcl = PointCloudList(xyz)
pcl.visualize_plotly()Or if you're dealing with point clouds with different numbers of points:
pcl = PointCloudList()
# If you haven't already created a `PointCloud` from the raw data:
for xyz, rgb in zip(xyzs, rgbs):
pcl.append_create_cloud(xyz, rgb)
# If you already have your data in the `PointCloud` format, you can do:
for pc in point_cloud_python_list:
pcl.append(pc)
# But if you already have multiple `PointCloud` objects in a raw Python list,
# you can directly initialize the `PointCloudList`:
pcl = PointCloudList(point_cloud_python_list)For further usage examples, look at the test functions.