Skip to content

Creating synthetic dataset for training object detection models

Notifications You must be signed in to change notification settings

Pendu/CarlaDatasetGeneration

Repository files navigation

Carla-synthetic-dataset-generation

Creating synthetic dataset for training the 3D bbox module.

Backbone from https://github.com/AlanNaoto/carla-dataset-runner

Setup prerequisites

  • Step1: Python carla package and its dependencies. Follow this guide to install Carla via github.
 https://carla.readthedocs.io/en/latest/start_quickstart/
  • Step2: Clone this repo. For other dependencies (to run data capturing scripts)
 pip install -r requirements.txt
  • Step3: Setup the python path for the carla egg file. Open the settings.py file and change the carla egg path to your own.

Running

  • step1: Launch CarlaEU4 Navigate to the folder where carla's pre-compiled package is present:
  ./CarlaUE4.sh TownXX -opengl 
  • step2: Launch data capturing

    Navigate to the folder where the repo is cloned:

  python3 main.py -ve 100 -wa 110 

Data captured

Illustration of the data captured in a newly created 'Data' folder.

Parameters :

  Type,BBox,Dimensions,Rotation_Y_(Red_Theta),Alpha_(Blue_Theta),Center_(ground_projection) 
  
  car,[ 96 364 246 464],"[5.19923, 2.61502, 2.47685]",-200.03459930419922,-108.02967071533203,"['6.315044', '-25.302868', '-0.001832']"
  car,[647 383 696 403],"[3.718622, 1.792448, 1.535494]",-91.14862298965454,0.8563055992126465,"['86.270348', '4.308983', '-0.010692']"
  pedestrian,[827 402 846 433],"[0.68, 0.68, 1.3]",87.95936584472656,179.96429443359375,"['65.018478', '-144.177292', '8.336428']"
  pedestrian,[825 383 837 403],"[0.68, 0.68, 1.86]",9.286087036132812,101.291015625,"['223.223145', '133.068954', '2.079820']"

Image :

sample

About

Creating synthetic dataset for training object detection models

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages