Virtual Multi-View Synthesis

Improving pedestrian heading estimation with Virtual Multi-View Synthesis (VMVS)

If you find this code useful, please consider citing:

@article{ku2019vmvs,
  title={Improving 3d Object Detection for Pedestrians with Virtual Multi-View Synthesis Orientation Estimation},
  author={Ku, Jason and Pon, Alex D and Walsh, Sean and Waslander, Steven L},
  journal={IROS},
  year={2019}
}

Setup

• Download the KITTI 3D Object Detection Dataset
• Go to scene_vis and generate dense depth maps
  • scene_vis/scripts/depth_completion/save_depth_maps_obj.py
• Clone this repo

git clone git@github.com:kujason/vmvs.git
pip install -r requirements

Demos

• save_crops_roi.py - Saves ROI crops of all pedestrians in the dataset (for comparison)
• save_crops_vmvs.py - Saves images from virtual views of all pedestrians in the dataset
• bash save_crops_vmvs_multiproc.sh - Runs multiple processes to save images
• show_crops_grid.py - Shows virtual views of pedestrians in a grid using multiple renderers (faster than save_crops_vmvs.py)