Abstract

We present DRACO, a method for Dense Reconstruction And Canonicalization of Object shape from one or more RGB images. Canonical shape reconstruction— estimating 3D object shape in a coordinate space canonicalized for scale, rotation, and translation parameters—is an emerging paradigm that holds promise for a multitude of robotic applications. Prior approaches either rely on painstakingly gathered dense 3D supervision, or produce only sparse canonical representations, limiting real-world applicability. DRACO performs dense canonicalization using only weak supervision in the form of camera poses and semantic keypoints at train time. During inference, DRACO predicts dense object-centric depth maps in a canonical coordinate-space, solely using one or more RGB images of an object. Extensive experiments on canonical shape reconstruction and pose estimation show that DRACO is competitive or superior to fully-supervised methods.

Downloads





Results

Given one or more views of an object, DRACO recovers a dense canonical reconstruction that are canonicalized for 6D pose and size estimation.

Input Image

Predicted Canonical NOCS Map

Canonical Reconstruction

Textured Canonical Reconstruction

MultiView Aggregation

Citation

BibTeX, 1 KB

@misc{sajnani2020draco,
title={DRACO: Weakly Supervised Dense Reconstruction And Canonicalization of Objects}, 
author={Rahul Sajnani and AadilMehdi Sanchawala and Krishna Murthy Jatavallabhula and Srinath Sridhar and K. Madhava Krishna},
year={2020},
eprint={2011.12912},
archivePrefix={arXiv},
primaryClass={cs.CV}
}

Acknowledgments

To be decided

Contact

For questions and clarifications please contact:
AadilMehdi Sanchawala
aadilmehdi.s@students.iiit.ac.in
Rahul Sajnani
rahul.sajnani@research.iiit.ac.in