3D Reloaded: Novel Algorithms for 3D Shape Inference and Analysis

Informations projet

3D Reloaded

N° de convention de subvention: 649323

DOI

10.3030/649323

Projet clôturé

Date de signature de la CE 5 Mai 2015

Date de début 1 Septembre 2015

Date de fin 31 Août 2020

Financé au titre de

• EXCELLENT SCIENCE - European Research Council (ERC)

Coût total

€ 2 000 000,00

Contribution de l’UE

€ 2 000 000,00

Coordonné par

TECHNISCHE UNIVERSITAET MUENCHEN
Germany

Periodic Reporting for period 4 - 3D Reloaded (3D Reloaded: Novel Algorithms for 3D Shape Inference and Analysis)

Période du rapport: 2020-03-01 au 2020-08-31

Résumé du contexte et des objectifs généraux du projet

The isssue adddressed in the project "3D Reloaded" is the development of novel algorithms for 3D reconstruction from cameras and for 3D shape analysis. Both of these problems are of central importance for machines to understand the 3D world observed with a camera.

This is important for society as it affects a number of application domains: For the development of self-driving cars, the autonomous cars needs to understand what happens around (and in particular) in front of the car. For augmented reality, one is interested in recovering the 3D structure of our world fIn the development of medical image analysis, algorithms need to interpret and analyze changes of inner organs represented in form of 3D shapes.

The overall objectives of the project are:
1. the development of algorithms for reconstructing 3D structures from monocular and from RGB-D cameras with a particular focus on dense and realtime capable methods,
2. the development of algorithms for 3D shape analysis, including problems such as measuring shape similarity, matching of shapes and the semantic analysis of shapes.
3. For the next phase of the project, a central objective is the development of 3D shape priors for 3D reconstruction.

Travail effectué depuis le début du projet jusqu’à la fin de la période considérée dans le rapport et principaux résultats atteints jusqu’à présent

We developed a number of algorithms for:
- reconstructing 3D scenes from monocular cameras with known pose,
- reconstrucing 3D scenes from RGB-D cameras with known pose,
- jointly reconstructing 3D scenes and camera trajectory from moving monocular cameras (visual odometry / simultaneous localization and mapping),
- jointly reconstructing 3D scenes and camera trajectory from moving RGB-D cameras,
- geometric shape priors for 3D reonstruction.

Progrès au-delà de l’état des connaissances et impact potentiel prévu (y compris l’impact socio-économique et les conséquences sociétales plus larges du projet jusqu’à présent)

The algorithms we developed were in many ways beyond the state of the art - in terms of accuracy, in terms of robustness, in terms of scalability, in terms of runtime, etc.

The main emphasis and expected results until the end of the project is the development of shape priors for 3D reconstruction. This is one of the topics we are currently working on.

Fight ill-posedness with ill-posedness: Single-shot variational, Haefner et al., CVPR 2018.

Divergence-free Shape Interpolation and Correspondence, Eisenberger, Laehner, Cremers, CVPR 2019.

Direct Sparse Odometry, Engel, Koltun, Cremers, PAMI 2018.