RANSAC in 2020: A CVPR Tutorial

Abstract

The main objective of this tutorial is to present the latest developments in robust model fitting. The tutorial will show the recent advancements in all three lines of research, including new sampling and local optimization methods in the traditional approach, novel branch-and-bound and mathematical programming algorithms in the global methods, and latest developments in differentiable alternative to RANSAC.

Organizers

Daniel Barath

Dmytro Mishkin

Rene Ranftl

Tat-Jun Chin

Ondra Chum

Jiri Matas

Introduction
Official time
9.00 - 9.45 PDT
Presenter
Jiri Matas
Short description
Introduction. The formulation and taxonomy of robust model estimation. Example problems. Outline of the Tutorial.
Presentation
Traditional approaches
Official time
9.45 - 10.45 PDT
Presenter
Ondra Chum
Short description
Traditional approaches for robust model fitting. Modules of the USAC framework.
Presentation
Latest developments in RANSAC
Official time
11.00 - 12.30 PDT
Presenter
Daniel Barath
Short description
The latest developments in the traditional RANSAC-like approaches.
Presentation
Mathematical programming approaches
Official time
13.30 - 15.00 PDT
Presenter
Tat-Jun Chin
Short description
Mathematical programming approaches including globally optimal algorithms (branch-and-bound, fixed-parameter tractable algorithms, etc.), deterministic refinement techniques, and preprocessing methods.
Presentation
Differentiable approaches
Official time
15.15 - 16.30 PDT
Presenter
Rene Ranftl
Short description
The latest developments in differentiable approaches for robust estimation.
Presentation
Experiments
Official time
16.30 - 17.30 PDT
Presenter
Dmytro Mishkin
Short description
A thorough experimental comparison of the state-of-the-art methods on fundemantel and essential matrix, PnP, homography and rigid transformation estimation.
Presentation