SIPAL-KITTI

Spatial Image Projection Autoencoder for LiDAR data

Project Structure

autoencoder

• main.py: Main script to train the autoencoder model.
• model.py: Defines the architecture of the autoencoder model used for compression.
• test.py: Contains testing functions for evaluating the model.

tools

• LiDAR_image_visualizer.py: Script to visualize LiDAR data projections and compressed images.
• projection_generator.py: Converts 3D LiDAR data into 2D images by projecting point clouds.

1. LiDAR projection generator

The LiDAR data from KITTI dataset is projected to 2D image, in size of 720 x 16, considering the FOV of Velodyne HDL-64E, which was used for data collection in KITTI datset. This code can be used for pointclouds from other LiDARs, with adjustment on FOV and number of points.

2. Autoencoder

The autoencoder implemented here is strongly affected by the delora repository. You can checkout the details in the linked repository.

How to use

1. Run pip install requirements.txt for installation of required python libraries.
2. Change path/to/your/directories in the codes to your desired directories.
3. Run LiDAR_image_visualizer.py to visualize your LiDAR data and run projection_generator.py to save it as png files.
4. Run main.py to train your autoencoder.
5. Run test.py to compare the original projection image and reconstructed version of it.