ROS2 Node that subscribes to PointCloud2 messages and visualizes them using Open3D.

pcd_subscriber_node.py

import sys
import os

import rclpy 
from rclpy.node import Node
import sensor_msgs.msg as sensor_msgs


import numpy as np
import open3d as o3d

class PCDListener(Node):

    def __init__(self):
        super().__init__('pcd_subsriber_node')

        ## This is for visualization of the received point cloud.
        self.vis = o3d.visualization.Visualizer()
        self.vis.create_window()
        self.o3d_pcd = o3d.geometry.PointCloud()


        # Set up a subscription to the 'pcd' topic with a callback to the 
        # function `listener_callback`
        self.pcd_subscriber = self.create_subscription(
            sensor_msgs.PointCloud2,    # Msg type
            'pcd',                      # topic
            self.listener_callback,      # Function to call
            10                          # QoS
        )

                
    def listener_callback(self, msg):
        # Here we convert the 'msg', which is of the type PointCloud2.
        # I ported the function read_points2 from 
        # the ROS1 package. 
        # https://github.com/ros/common_msgs/blob/noetic-devel/sensor_msgs/src/sensor_msgs/point_cloud2.py

        pcd_as_numpy_array = np.array(list(read_points(msg)))
        

        # The rest here is for visualization.
        self.vis.remove_geometry(self.o3d_pcd)
        self.o3d_pcd = o3d.geometry.PointCloud(
                            o3d.utility.Vector3dVector(pcd_as_numpy_array))
        self.vis.add_geometry(self.o3d_pcd)
        self.vis.poll_events()
        self.vis.update_renderer()

        


## The code below is "ported" from 
# https://github.com/ros/common_msgs/tree/noetic-devel/sensor_msgs/src/sensor_msgs
# I'll make an official port and PR to this repo later: 
# https://github.com/ros2/common_interfaces
import sys
from collections import namedtuple
import ctypes
import math
import struct
from sensor_msgs.msg import PointCloud2, PointField

_DATATYPES = {}
_DATATYPES[PointField.INT8]    = ('b', 1)
_DATATYPES[PointField.UINT8]   = ('B', 1)
_DATATYPES[PointField.INT16]   = ('h', 2)
_DATATYPES[PointField.UINT16]  = ('H', 2)
_DATATYPES[PointField.INT32]   = ('i', 4)
_DATATYPES[PointField.UINT32]  = ('I', 4)
_DATATYPES[PointField.FLOAT32] = ('f', 4)
_DATATYPES[PointField.FLOAT64] = ('d', 8)

def read_points(cloud, field_names=None, skip_nans=False, uvs=[]):
    """
    Read points from a L{sensor_msgs.PointCloud2} message.

    @param cloud: The point cloud to read from.
    @type  cloud: L{sensor_msgs.PointCloud2}
    @param field_names: The names of fields to read. If None, read all fields. [default: None]
    @type  field_names: iterable
    @param skip_nans: If True, then don't return any point with a NaN value.
    @type  skip_nans: bool [default: False]
    @param uvs: If specified, then only return the points at the given coordinates. [default: empty list]
    @type  uvs: iterable
    @return: Generator which yields a list of values for each point.
    @rtype:  generator
    """
    assert isinstance(cloud, PointCloud2), 'cloud is not a sensor_msgs.msg.PointCloud2'
    fmt = _get_struct_fmt(cloud.is_bigendian, cloud.fields, field_names)
    width, height, point_step, row_step, data, isnan = cloud.width, cloud.height, cloud.point_step, cloud.row_step, cloud.data, math.isnan
    unpack_from = struct.Struct(fmt).unpack_from

    if skip_nans:
        if uvs:
            for u, v in uvs:
                p = unpack_from(data, (row_step * v) + (point_step * u))
                has_nan = False
                for pv in p:
                    if isnan(pv):
                        has_nan = True
                        break
                if not has_nan:
                    yield p
        else:
            for v in range(height):
                offset = row_step * v
                for u in range(width):
                    p = unpack_from(data, offset)
                    has_nan = False
                    for pv in p:
                        if isnan(pv):
                            has_nan = True
                            break
                    if not has_nan:
                        yield p
                    offset += point_step
    else:
        if uvs:
            for u, v in uvs:
                yield unpack_from(data, (row_step * v) + (point_step * u))
        else:
            for v in range(height):
                offset = row_step * v
                for u in range(width):
                    yield unpack_from(data, offset)
                    offset += point_step

def _get_struct_fmt(is_bigendian, fields, field_names=None):
    fmt = '>' if is_bigendian else '<'

    offset = 0
    for field in (f for f in sorted(fields, key=lambda f: f.offset) if field_names is None or f.name in field_names):
        if offset < field.offset:
            fmt += 'x' * (field.offset - offset)
            offset = field.offset
        if field.datatype not in _DATATYPES:
            print('Skipping unknown PointField datatype [%d]' % field.datatype, file=sys.stderr)
        else:
            datatype_fmt, datatype_length = _DATATYPES[field.datatype]
            fmt    += field.count * datatype_fmt
            offset += field.count * datatype_length

    return fmt



def main(args=None):
    # Boilerplate code.
    rclpy.init(args=args)
    pcd_listener = PCDListener()
    rclpy.spin(pcd_listener)
    
    # Destroy the node explicitly
    # (optional - otherwise it will be done automatically
    # when the garbage collector destroys the node object)
    pcd_listener.destroy_node()
    rclpy.shutdown()

if __name__ == '__main__':
    main()

Wow, it's very detailed and well made, thank you!!
it really helped me.

I've got a very big point cloud and the o3d.utility.Vector3dVector line is throwing a runtime error without any description when using your sample above.

The following code works though. Sample from https://answers.ros.org/question/255351/how-o-save-a-pointcloud2-data-in-python/

        gen = read_points(msg, skip_nans=True)
        int_data = list(gen)

        xyz = np.empty((len(int_data), 3))
        rgb = np.empty((len(int_data), 3))
        idx = 0
        for x in int_data:
            test = x[3] 
            # cast float32 to int so that bitwise operations are possible
            s = struct.pack('>f' ,test)
            i = struct.unpack('>l',s)[0]
            # you can get back the float value by the inverse operations
            pack = ctypes.c_uint32(i).value
            r = (pack & 0x00FF0000)>> 16
            g = (pack & 0x0000FF00)>> 8
            b = (pack & 0x000000FF)
            # prints r,g,b values in the 0-255 range
                        # x,y,z can be retrieved from the x[0],x[1],x[2]
            # xyz = np.append(xyz,[[x[0],x[1],x[2]]], axis = 0)
            # rgb = np.append(rgb,[[r,g,b]], axis = 0)
            xyz[idx] = [x[0],x[1],x[2]]
            rgb[idx] = [r,g,b]
            idx = idx + 1

        out_pcd = o3d.geometry.PointCloud()    
        out_pcd.points = o3d.utility.Vector3dVector(xyz)
        out_pcd.colors = o3d.utility.Vector3dVector(rgb)

Any idea why the numpy_array returned from your code sample would not work?

I solved the error changing this line in read_points()
yield unpack_from(data, offset)[:3]
It because this line was returning lists of 4 positions and o3d.utility.Vector3dVector works with lists of 3 positions.
Note: I don't know for what the 4 position value is for. Maybe it is the point color packed in float.

Thanks for the solution!