Making Machine Learning on Static and Dynamic 3D Data Practical
Final Report Abstract
This project was a collaborative effort between NiessnerLab at the Technical University of Munich and researchers at the Skolkovo Institute of Science and Technology. The project aimed to develop advanced machine learning algorithms to automate the creation and annotation of 3D indoor scenes. Over its duration, the project achieved significant milestones, resulting in numerous research publications and contributing to the academic and industrial communities. The project made significant progress in the following areas: Development of novel machine learning algorithms for 3D scene generation. - Creation and annotation of large datasets of 3D indoor scenes. - Extensive experimental validation and refinement of the developed methods. - Dissemination of findings through high-impact publications.
Publications
-
"CAD-Deform: Deformable Fitting of CAD Models to 3D Scans". ECCV 2020
Vladislav Ishimtsev et al.
-
"Pri3D: Can 3D Priors Help 2D Representation Learning?” ICCV’21
Ji Hou et al.
-
DEF: Deep Estimation of Sharp Geometric Features in 3D Shapes. ACM ToG’22
Albert Matveev et al.
-
“4DContrast: Contrastive Learning with Dynamic Correspondences for 3D Scene Understanding” ECCV’22
Chen et al.
-
“D3Net: A Speaker-Listener Architecture for Semi-supervised Dense Captioning and Visual Grounding in RGB-D Scans” ECCV’22
Chen et al.
-
“Pose2Room: Understanding 3D Scenes from Human Activities” ECCV’22
Nie et al.
-
“RC-MVSNet: Unsupervised Multi-View Stereo with Neural Rendering” ECCV’22
Chang et al.
-
“Texturify: Generating Textures on 3D Shape Surfaces” ECCV’22
Siddiqui et al.
-
"ObjectMatch: Robust Registration using Canonical Object Correspondences" CVPR’23
Gümeli et al.
-
"Synthetic Data for Improved 3D Scene Generation." CVPR’23
Nie et al.
-
“Deep vectorization of technical drawings" Springer’23
Vage Egiazarian et al.
-
“Learning Neural Parametric Head Models” CVPR’23
Giebenhain et al.
