Skip to content

Implement Nvidia AI Lab's paper "Towards Optimal Strategies for Training Self-Driving Perception Models in Simulation" based on "Lift, Splat, Shoot"

License

Notifications You must be signed in to change notification settings

XinWang0219/Sim2Real

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Sim2Real

Implement Nvidia AI Lab's paper "Towards Optimal Strategies for Training Self-Driving Perception Models in Simulation" based on "Lift, Splat, Shoot"

Project Structure

  • carla_data_generation Generate nuScence style like Carla dataset including six cameras around the ego vehicle, Lidar, Town03 maps, and annotations of pedestrain and vehicles.

  • style_transform Use MUNIT style transform baseline to train synthesis to cityscape style transform model with Carla dataset and NuScene mini dataset.

  • lift_splat_shoot Based on the "Lift Splat Shoot" BEV segmentation model, train lift_splat model with original Carla dataset and lift_splat_adpat model with style transformed Carla dataset from scratch.

Comparing the performance of lift_splat_adpat model and lift_splat model on target domain (NuScene) to certificate the base theory, "domain adaption", is a way to implement Sim2Real.

Project Dataset

Project Result

Project Analysis

Reference Announcement

  1. The paper "Towards Optimal Strategies for Training Self-Driving Perception Models in Simulation", https://arxiv.org/abs/2111.07971
  2. The paper and source code "Lift, Splat, Shoot: Encoding Images From Arbitrary Camera Rigs by Implicitly Unprojecting to 3D", https://github.com/nv-tlabs/lift-splat-shoot
  3. Carla Dataset Generator: https://github.com/cf206cd/carla_nuscenes
  4. MUNIT source code: https://github.com/NVlabs/MUNIT

About

Implement Nvidia AI Lab's paper "Towards Optimal Strategies for Training Self-Driving Perception Models in Simulation" based on "Lift, Splat, Shoot"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published