Introduction

Point-viz is a light-weight, out-of-box point cloud visualizer working in browser, it is built on Three.js and has python API. Bounding box (bbox) visualization is supported and here is an example. Point-viz does not need Internet connection to work (but the installation needs, of course), and it has been tested on Chrome and Safari with both python 2 and 3. The main motivation of this visualizer is to allow user to visualize point cloud data from remote server easily .

Tested Environment

• python==3.7
• numpy==1.18.5
• Browsers: Google Chrome, Safari, Firefox

Usage

To install point-viz, simply run: pip install point-viz, and that's all.

Below is the template of its python API:

# Import package.
from point_viz import PointvizConverter

# Initialize and setup output directory.
Converter = PointvizConverter(home)
# Pass data and create html files.
Converter.compile(task_name, coors, default_rgb, intensity, bbox_params, bbox_color)

Below is a more specific (partial) example that is demonstrated on KITTI dataset.

save_viz_path = /path/to/save/visualization/

Converter = PointvizConverter(save_viz_path)

...

# note the pts_inputs are in z-up system
# which in kittidataset, it should be in 
# velodyne coordinates
pts_inputs = kitti_dataset.get_sample(1)

...

gt_bbox_params_list = []

# the labels are in z-up system
# which in kittidataset, it should be in 
# velodyne coordinates
for obj in kitti_dataset.get_labels(1):
    bbox_params_list = [obj.l, # along y axis
                        obj.h, # along z axis
                        obj.w, # along x axis
                        obj.y,
                        obj.z,
                        obj.x,
                        obj.ry] 
                        # in l, h, w, y, z, x
                        # since the coordinate system in three.js is y-up
    gt_bbox_params_list.append(bbox_params_list)

# if you want to generate visualizations for multiple samples
# you have to call this function compile in a loop
Converter.compile("sample_{}".format(1), coors=pts_input[:,:3],         intensity=pts_input[:,[1,2,0]], bbox_params=gt_bbox_params_list)

In the example above you find the following directory architectures being generated:

• /path/to/save/visualization/
  • data
    • sample_1_bbox.csv (if bbox_params is not None during compilation)
    • sample_1_pc.bin (numpy binary files that contains x, y, z, r, g, b)
    • ... (more if you compile multiple times)
  • html
    • sample_1.html
    • ... (more if you compile multiple times)
  • src
    • containing the dependencies that html requires

In your remote server, cd /path/to/save/visualization/. Then run python3 -m http.server 8000.

To access the visualization locally, go to the url http://<ip_address_of_the_remote_server>:8000

Variables Explanation

The * means optional.

home: The directory where to put output html files, must be given.
task_name*: string, name of the output html file (can be overwritten if the name already exists; default value is "default").
coors*: 2-D float array, the x, y and z coordinates of each point in the point cloud.
default_rgb*: 2-D float/int array of the same length as coors, the R, G and B colors of each point. If not provided, the RGB will be automatically calculated based on intensity (if given) or point coordinates (when intensity is also missing).
intensity*: 1-D float array of the same length as coors, the intensity of each point. It only takes effect when default_rgb is not given.
bbox_params*: 2-D list, the geometry parameters of each bbox. Attributes of each row should be arranged as follows:

bbox_color*: boolean, default is True. If the color of bbox is missing while the label text is given, then bbox_color has to be explicitly set to False.