Stochastic Interpolants for Controllable Scene Generation

Stochastic interpolants offer a powerful toolbox for generative modeling, including denoising diffusion, flow matching, and inductive moment matching. In the context of ETHAR, they hold great promise as a path towards controllable sampling and synthesis of multimodal content across the 2D and 3D domains. This potential is exemplified by scenarios where users aim to reimagine or enhance physical environments based on minimal imagery and high-level guidance. The proposed framework leverages stochastic interpolants to generate semantically consistent and geometrically plausible 3D reconstructions, including novel views that complete and enrich the scene beyond what is directly observable. We propose to develop a framework for interactive, semantically guided, and geometrically coherent content creation based on stochastic interpolants with variational regularization, in service of immersive user experiences and augmented reality.

Project Team

Prof. Konrad Schindler

Konrad Schindler is a Full Professor in the Department of Civil, Environmental and Geomatic Engineering at ETH Zurich, where he leads the Institute of Geodesy and Photogrammetry. His research primarily focuses on photogrammetry, remote sensing, computer vision, and image understanding, driving advancements in these critical fields. His work contributes significantly to the scientific community, enhancing our understanding of geospatial data and imaging technologies.

Dr. Dominik Narnhofer

Dr. Dominik Narnhofer is a Computer Vision/Machine Learning Researcher serving as a senior researcher at the Photogrammetry and Remote Sensing Group (D-BAUG) at ETH Zurich.

PD Dr. Federico Tombari

Federico Tombari is a Research Director at Google Zurich, Switzerland, where he leads an applied research team in Computer Vision and Machine Learning across the US, Switzerland, and Germany. He is also affiliated with the Faculty of Computer Science at TUM as a lecturer (PrivatDozent) at the CAMP Chair. An up-to-date list of publications is available at his Google Scholar. Federico Tombari's research activity is focused on different aspects of computer vision and machine learning, with a focus on 3D computer vision (e.g. scene understanding, 3D object recognition, 3D reconstruction, SLAM). The fields of application of his research are mainly in robotics, augmented reality, autonomous driving and healthcare. He is currently particularly excited about unsupervised learning for visual data, Large Multimodal Models, (Neural) Radiance Fields, and scene graphs for scene understanding.

Publications

Understanding, Accelerating, and Improving MeanFlow Training

Authors:
Jin-Young Kim
Hyojun Go
Lea Bogensperger
Julius Erbach
Nikolai Kalischek
PD Dr. Federico Tombari
Prof. Konrad Schindler
Dr. Dominik Narnhofer
Published:
to be published at CVPR, 2026

Generative, panoramic depth estimation (working title)

Published:
submitted to NeurIPS, 2026

VIST3A: Text-to-3D by Stitching a Multi-View Reconstruction Network to a Video Generator

Authors:
Hyojun Go
Dr. Dominik Narnhofer
Goutam Bhat
Prune Truong
PD Dr. Federico Tombari
Prof. Konrad Schindler
Published:
ICLR 2026 (Oral)