Simple Kitti Visualizations

Simple and easy to follow point cloud and bounding box visualizations for kitti data. The only dependencies are OpenCV and Numpy.

Create re-projections onto images and 3 different views of a point-cloud with a black background (birds eye view, first person view, third person view).

Installation

Clone the repository.

Use the package manager pip to install the requirements.

pip install -r requirements.txt

Usage

In main.py, change the filepaths to the location of your kitti files. Run vis_image or vis_points on your data.

example:

# define frame name and get path for each file
frame = '000032'
img_path = 'data/' + frame + '.png'
pts_path = 'data/' + frame + '.bin'
calib_path = 'data/' + frame + '_calib.txt'
label_path = 'data/' + frame + '_label.txt'

# define output path for image visualisation and create image
output_path = "output/vis_image.png"
img = vis_image(
    output_path,
    img_path,
    pts_path,
    calib_path,
    label_path,
    lidar=False,
    boxes_2d=False,
    boxes_3d=True
)

cv2.imwrite(output_path, img)
print("Saved img:", output_path)

Results

3D Bounding Boxes on Image

2D Bounding Boxes on Image

Point Cloud on Image

Third Person View Point Cloud

First Person View Point Cloud

Birds Eye View Point Cloud