It may be helpful to look at the CARLA Skeleton page to see what skeleton is used to express the pose data. The data was generated using code provided in the repository, which is part of a bigger collection of repos.

Available versions

  • Binary Single Pedestrian
    Generated with a 1600×600 image size, it contains recorded video clips and semantic segmentation labels. It also provides crossing/non-crossing annotations. Clips are 900 frames long @ 30 FPS.

Old/internal use versions

  • Wide Camera Pedestrians
    Version with 1600×600 image size. It introduced more pedestrian behaviors but also resulted in many empty videos. The clips contained crossing labels but were very short (600 frames @ 30 FPS).
  • Basic Pedestrians Crossing
    The first version of the generated data. It used an 800×600 image size, and the pedestrian was often not visible in the video. It also didn’t contain crossing labels or semantic labels.