Dec 13, 2018 · This paper introduces SIGNet, a novel framework that provides robust geometry perception without requiring geometrically informative labels.

In SIGNet, we strive to achieve robust performance for depth and flow perception without using geometric labels. To achieve this goal, SIGNet utilizes semantic.

Abstract: Unsupervised learning for geometric perception (depth, optical flow, etc.) is of great interest to autonomous systems.

Specifically, SIGNet integrates semantic information to make unsupervised robust geometric predictions for objects in low lighting and noisy environments.

Semantic Instance Geometry Network for Unsupervised Percepetion. Project page: https://mengyuest.github.io/SIGNet/

SIGNet is shown to improve upon the state-of-the-art unsupervised learning for depth prediction by 30% (in squared relative error), and improves the dynamic ...

In SIGNet, we strive to achieve robust performance for depth and flow perception without using geometric labels. To achieve this goal, SIGNet utilizes semantic.

SIGNet is shown to improve upon the state-of-the-art unsupervised learning for depth prediction by 30% (in squared relative error). Ranked #67 on Monocular ...

"SIGNet: Semantic Instance Aided Unsupervised 3D Geometry Perception", (CVPR), 2019. [arXiv pdf]. This repository recreates SIGNet for the CARLA Simulator.

Signet: Semantic instance aided unsupervised 3d geometry perception. Y Meng, Y Lu, A Raj, S Sunarjo, R Guo, T Javidi, G Bansal, D Bharadia. Proceedings of the ...