An Investigation of Unsupervised Cell Tracking and Interactive Fine-Tuning.
Bio Image Computing Workshop, International Conference for Computer Vision (ICCV) (2025).
We introduce Attrackt—an unsupervised loss for cell tracking and combine it with active learning and low-rank fine-tuning to enable annotation-efficient, interactive model refinement.

Unsupervised Learning of Object-Centric Embeddings for Cell Instance Segmentation in Microscopy Images.
International Conference for Computer Vision (ICCV) (2023).
DOI: 10.48550/arXiv.2310.08501
We introduce Cellulus—an unsupervised method for cell instance segmentation, where the core idea is to predict relative spatial offsets between image patches, thus learning object-centric embeddings, enabling cell segmentation and reducing the need for manual annotations.
GitHub

EmbedSeg: Embedding-based Instance Segmentation for Biomedical Microscopy Data.
Medical Image Analysis (2022).
DOI: 10.1016/j.media.2022.102523
We present EmbedSeg—a supervised method for cell instance segmentation, which achieves high accuracy while maintaining a low GPU memory footprint.
Slides GitHub

Registration of Multi‐modal Volumetric Images by Establishing Cell Correspondence.
Bio Image Computing Workshop, European Conference for Computer Vision (ECCV) (2020).
DOI: 10.1007/978-3-030-66415-2_30
We present PlatyMatch—a method for volumetric image registration, which establishes cell-to-cell correspondences across two microscopy modalities, and uses these matches to register volumetric images.
Slides GitHub

Fully Unsupervised Probabilistic Noise2Void
International Symposium on Biomedical Imaging (ISBI) (2020).
DOI: 10.48550/arXiv.1911.12291
We present PPN2V—an unsupervised method for image denoising, which requires only individual noisy microscopy images and no calibration data.
Slides GitHub