WareSynth

A procedural synthetic data generation pipeline that can be used to generate 3D warehouse scenes and automate the process of data capture and generating corresponding annotation.

A Sample Warehouse

Depth and Trajectory Example

WareSynth Synthetic Warehouse Generation Pipeline is a part of the project: “ Monocular Multi-Layer Layout Estimation for Warehouse Racks “

Link to arxiv Paper, Monocular Multi-Layer Layout Estimation for Warehouse Racks

Link to Dataset generated using WareSynth (used for training network)

Videos

Capabilities

Capture camera positions along with the intrinsic and extrinsic parameters for each view.
Extract 2D annotations such as 2D bounding boxes and semantic and instance segmentation masks of the objects.
Obtain the 3D positions, orientations and 3D bounding boxes for all objects present in the camera FOV, along with depth maps and normal information.
Our pipeline can also be used to obtain stereo-information.
The annotations can be exported in various popular formats like COCO, YOLO, Pix3D, KITTI, BOP as needed by the user.

Depth maps and camera trajectory

Depth and Trajectory Example

2D Instance segmentation masks

Depth and Trajectory Example

3D bounding boxes,shown on point cloud generated from RGB-D data

Depth and Trajectory Example

How to cite

If you find this work useful, please use the following BibTeX entry for citing us

@article{nigam2021racklay,
  title={RackLay: Multi-Layer Layout Estimation for Warehouse Racks},
  author={Nigam, Meher Shashwat and Prabhu, Avinash and Sahu, Anurag and Gupta, Puru and Karandikar, Tanvi and Shankar, N Sai and Sarvadevabhatla, Ravi Kiran and Krishna, K Madhava},
  journal={arXiv preprint arXiv:2103.09174},
  year={2021}
}

Contact information

anurag.sahu@research.iiit.ac.in
tanvi.karandikar@students.iiit.ac.in