• geometry
• intensity
• labels
• color
• mask

• geometry
• intensity
• labels
• color
• mask

GTA plugins are available in directory GTA. Please follow the instructions from the original GTAVisionExport repository to make it work.

Some notable changes from the original repository:

• 4 cameras which switch themselves sitting atop a car to simulate lidar, each with 65 VFOV (~91 HFOV)
• It is made substantially more lightweight in order to keep only necessary parts.

First go through GTA directory to setup the GTA plugins and collect dataset

Then go through python directory, to compute velodyne-like points.

In order to predict intensity run 'python/infer_intensity.py' with first argument with path to source folder, where grid lidar sweeps are, and second argument output directory, where you want your point clouds with intensity stored. Point clouds will have channels: Depth, X, Y, Z, Intensity, Label, Red, Green, Blue, Color_mask, Returned_ray_mask

Folder configs contains configs for starting either python/model_eval.py or python/model_train.py. Each config in the toplevel configs directory is a list of configs from includes, which are merged together. The order is important, if there are two same keys in different included files, the one that was included later is kept.

You will probably want to change configs/includes/{gta,kitti}data/basedata.yml and config/includes/{gta,kitti}data/segment_data.yml, which contains paths for data.

Configs configs/includes/{reflect,segment}_configs/model*.yml contain directories, where checkpoints are stored. You probably want to change that too.