Home

Awesome

ConOR: Uncertainty Quantification in Depth Estimation via Constrained Ordinal Regression

This repository contains the implementation of ConOR on synthetic data and benchmark dataset described in D. Hu, L. Peng, T. Chu, X. Zhang, Y. Mao, H. Bondell, and M. Gong: Uncertainty Quantification in Depth Estimation via Constrained Ordinal Regression. European Conference on Computer Vision (ECCV) 2022.

Supplementary material can be found here.

Input ImageDepth Estimation
0.50.5
Predictive UncertaintyEstimation Error (vs. ground truth)
0.50.5

Synthetic Data for Demonstration

The Simulation directory contains the code of the experiment on synthetic data. To run the experiment, please refer to scipts.

Experimental Results on Benchmark

The Benchmark directory contains the code of the experiment on KITTI and NYUv2. To run the experiment, please refer to scipts.

Citation

If you find it useful, please consider citing:

@inproceedings{hu2022uncertainty,
  title={Uncertainty Quantification in Depth Estimation via Constrained Ordinal Regression},
  author={Hu, Dongting and Peng, Liuhua and Chu, Tingjin and Zhang, Xiaoxing and Mao, Yinian and Bondell, Howard and Gong, Mingming},
  booktitle={Computer Vision--ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23--27, 2022, Proceedings, Part II},
  pages={237--256},
  year={2022},
  organization={Springer}
}

Acknowledgement

The benchmark code base is origined from an awesome DORN pytorch implementation.