Florian Hofherr

PhD student @ TUM CVG

me.jpg

I am a PhD student in the Computer Vision Group at the Technical University of Munich (TUM), under the supervision of Prof. Daniel Cremers. My research interests focus on neural fields and their applications to various challenges such as geometry representation, dynamic scene reconstruction, and material modeling. More recently, I have gained hands-on experience with diffusion models in an industrial research setting as an Applied Scientist Intern at Zillow.

I hold an M.Sc. in Mathematics in Science and Engineering, with a specialization in control theory and flight system dynamics, and a B.Sc. in Engineering Science, both from the Technical University of Munich. I also spent semesters abroad at the Federal Institute of Technology Zürich (ETH) and the University of Queensland.



Publications

2026

  1. figueroa2026diamond.png
    DIAMOND-SSS: Diffusion-Augmented Multi-View Optimization for Data-efficient SubSurface Scattering
    Guillermo Figueroa-Araneda, Iris Diana Jimenez, Florian Hofherr, Manny Ko, Hector Andrade-Loarca, and Daniel Cremers
    arXiv preprint, 2026
    arXiv


    We use diffusion models for novel-view synthesis and relighting to augment sparse multi-view captures for relightable 3D reconstruction. We reconstruct the scene using Gaussian splatting with a subsurface scattering extension and introduce loss terms to correct inconsistencies.

2025

  1. hofherr2025neuralBRDFs.jpg
    On Neural BRDFs: A Thorough Comparison of State-of-the-Art Approaches
    Florian Hofherr, Bjoern Haefner, and Daniel Cremers
    WACV, 2025 (Oral Presentation)
    Project Page arXiv Video


    We perform an extensive comparison of different neural BRDF models. Moreover, we propose a novel input mapping that ensures reciprocity by construction and an enhancement of neural BRDFs based on an additive split architecture.

  2. saroha2025zdyss.png
    ZDySS–Zero-Shot Dynamic Scene Stylization using Gaussian Splatting
    Abhishek Saroha, Florian Hofherr, Mariia Gladkova, Cecilia Curreli, Or Litany, and Daniel Cremers
    arXiv preprint, 2025
    arXiv


    We propose a zero-shot approach for stylizing dynamic 3D scenes that operates on feature-augmented Gaussian splatting representations, enabling spatio-temporally consistent results across views and time.

2024

  1. mahajan2024meshfeat.jpg
    MeshFeat: Multi-Resolution Features for Neural Fields on Meshes
    Mihir Mahajan*, Florian Hofherr*, and Daniel Cremers
    ECCV, 2024
    Project Page arXiv Video


    We adapt the idea of parametric multi-resolution feature grid encodings from Euclidean space to meshes. This enables neural fields with smaller MLPs resulting in a significant evaluation speedup.

2023

  1. real_pendulum.gif
    Neural Implicit Representations for Physical Parameter Inference from a Single Video
    Florian Hofherr, Lukas Koestler, Florian Bernard, and Daniel Cremers
    WACV, 2023 (Spotlight Presentation)
    Project Page arXiv Video


    We integrate implicit representations for appearance with explicit physical dynamics models to reconstruct dynamic scenes. This approach not only enables accurate reconstruction but also supports physically realistic scene editing.

2018

  1. hofherr2018lagrangian.jpg
    Lagrangian Transport Through Surfaces in Compressible Flows
    Florian Hofherr, and Daniel Karrasch
    SIAM Journal on Applied Dynamical Systems, 2018
    arXiv