Tim Büchner

Tim Büchner

Computer Vision Group Jena, Research Associate, PhD Student

I am currently doing my PhD at the Computer Vision Group Jena at the Friedrich Schiller University Jena under Prof. Dr.-Ing. Joachim Denzler. My research brings physical and medical structure into data-driven models. The main areas are multi-modal facial analysis and physics-informed machine learning, with a supporting role in medical image segmentation.

multi-modal facial research

The human face is one of the most expressive parts of the body, and every movement comes from the mimic musculature, innervated by the facial nerve. Most computer vision studies the face through its visible surface alone, using images and video. We treat it as a multi-modal object instead, recording high-resolution 3D geometry and surface electromyography (sEMG) at the same time as video. This lets us relate muscle activation directly to the resulting expression rather than inferring one from the other. We reconstruct facial expressions both implicitly and explicitly in 3D, look at which facial properties expression classifiers actually rely on, and build visualizations that make muscle activity legible on the face itself.

Facial palsy is our clinical application. Injury to the facial nerve causes unilateral motor dysfunction, with consequences from incomplete eye closure to impaired mimicry. In the DFG project Bridging the Gap: Mimics and Muscles, we combine the 3D surface with the underlying muscle activity to understand facial expressions better. The project is a collaboration between the Computer Vision Group Jena and the Ear-Nose-Throat Clinic Jena under Prof. Dr. Orlando Guntinas-Lichius, bringing together computer vision and medical science to help patients with facial palsy.

physics-informed machine learning

Most physical systems carry a strong prior structure: governing equations, conservation laws, symmetries. Purely data-driven models ignore it. Physics-Informed Neural Networks (PINNs) make these constraints part of learning, so that models stay consistent with known science, especially where measurements are expensive or sparse. I work on architectural building blocks that address core limitations of PINNs in scalability and expressivity via functional tensor decompositions, and apply them to inverse problems.

sensorized surgery

In a supporting role, I contribute to the Sensorized Surgery project, which builds a sensor-equipped surgical system for head and neck tumor resection. It combines multi-modal marker-free imaging, mechanical tissue sensing, and AI-driven analysis to predict tumor boundaries in real time and give the surgeon visual and haptic feedback. I support my group in the semantic segmentation research that separates tumor from surrounding tissue, with a focus on staying robust under distribution shifts, small datasets, and noisy labels in the clinical setting.

Selected Publications

2025

  1. Büchner, T., Anders, C., Guntinas-Lichius, O., & Denzler, J. (2025). Electromyography-Informed Facial Expression Reconstruction for Physiological-Based Synthesis and Analysis. ArXiv Preprint ArXiv:2503.09556. https://doi.org/10.1109/CVPR52734.2025.00029
    scholar (5) html arXiv doi

2024

  1. Vemuri, S. K., Büchner, T., Niebling, J., & Denzler, J. (2024). Functional Tensor Decompositions for Physics-Informed Neural Networks. International Conference on Pattern Recognition (ICPR), 32–46. https://doi.org/10.1007/978-3-031-78389-0_3
    scholar (26) code arXiv doi
  2. Büchner, T., Penzel, N., Guntinas-Lichius, O., & Denzler, J. (2024, December). Facing Asymmetry - Uncovering the Causal Link between Facial Symmetry and Expression Classifiers using Synthetic Interventions. Asian Conference on Computer Vision (ACCV). https://doi.org/10.1007/978-981-96-0911-6_26
    scholar (4) arXiv doi

2023

  1. Büchner, T., Guntinas-Lichius, O., & Denzler, J. (2023). Improved Obstructed Facial Feature Reconstruction for Emotion Recognition with Minimal Change CycleGANs. In W. P. J. Blanc-Talon D. Popescu & P. Scheunders (Eds.), Advanced Concepts for Intelligent Vision Systems (Acivs) (pp. 262–274). SpringerNature. https://doi.org/10.1007/978-3-031-45382-3_22
    scholar (9) doi
  2. Büchner, T., Sickert, S., Volk, G. F., Anders, C., Guntinas-Lichius, O., & Denzler, J. (2023). Let’s Get the FACS Straight - Reconstructing Obstructed Facial Features. International Conference on Computer Vision Theory and Applications (VISAPP), 727–736. https://doi.org/10.5220/0011619900003417
    scholar (8) arXiv doi

News

Latest Posts