ComfyUI Node: PointCloudCleaner

PointCloudCleaner:

The PointCloudCleaner node is designed to refine and optimize point cloud data by projecting it through an identity matrix into a fisheye-180° normalized UV and depth space. This process involves inverting and normalizing the depth values and performing a voxel-based cleaning operation in the (px, py, inv_depth) space. The primary goal of this node is to enhance the quality of point cloud data by removing noise and redundant points, which can significantly improve the accuracy and efficiency of subsequent processing tasks. By leveraging voxel-based cleaning, the node ensures that only the most relevant and densely populated areas of the point cloud are retained, making it an essential tool for applications that require high-quality 3D data, such as 3D modeling, virtual reality, and augmented reality.

PointCloudCleaner Input Parameters:

pointcloud

The pointcloud parameter is a tensor that represents the 3D point cloud data to be processed. This data is crucial as it forms the basis of the cleaning operation, where each point in the cloud is evaluated and potentially filtered based on its spatial characteristics.

width

The width parameter specifies the width of the output image in pixels. It influences the resolution of the projected point cloud and can range from 1 to 16384, with a default value of 1024. A higher width value results in a more detailed projection, which can be beneficial for applications requiring high precision.

height

The height parameter defines the height of the output image in pixels, similar to the width parameter. It also ranges from 1 to 16384, with a default value of 1024. Adjusting the height affects the vertical resolution of the projection, allowing for more detailed vertical features in the point cloud.

voxel_size

The voxel_size parameter determines the size of the voxels used in the cleaning process. It is a floating-point value with a minimum of 0.001 and a default of 1.0. Smaller voxel sizes result in finer granularity, allowing for more precise cleaning, while larger sizes may speed up processing by grouping more points together.

min_points_per_voxel

The min_points_per_voxel parameter sets the minimum number of points required within a voxel for it to be retained in the cleaned point cloud. This integer parameter has a minimum value of 1 and a default of 3. Increasing this value can help eliminate noise by ensuring that only densely populated areas are kept.

PointCloudCleaner Output Parameters:

cleaned_pointcloud

The cleaned_pointcloud is the output tensor that contains the refined point cloud data after the cleaning process. This output is crucial as it represents the optimized version of the input point cloud, with noise and redundant points removed, making it more suitable for further processing or analysis.

PointCloudCleaner Usage Tips:

• To achieve the best results, adjust the voxel_size and min_points_per_voxel parameters based on the density and noise level of your input point cloud. Smaller voxel sizes and higher minimum points per voxel can help in retaining only the most relevant data.
• Experiment with different width and height values to find the optimal resolution for your specific application. Higher resolutions can provide more detail but may require more computational resources.

PointCloudCleaner Common Errors and Solutions:

ValueError: No waypoints recorded. Please record at least one with 'P'.

• Explanation: This error occurs when there are no waypoints recorded for the trajectory, which is necessary for interpolation.
• Solution: Ensure that you have recorded at least one waypoint using the 'P' key before attempting to process the point cloud.

RuntimeError: CUDA out of memory

• Explanation: This error indicates that the GPU does not have enough memory to process the point cloud with the current settings.
• Solution: Try reducing the width, height, or voxel_size parameters to decrease the memory requirements, or consider using a machine with more GPU memory.