Robert Kiewisz
I build machine-learning tools that turn electron-microscopy images into measurable structure — automating the segmentation and reconstruction work that used to be done by hand, one slice at a time.
Focus
Computer vision and image processing for structural biology: deep-learning segmentation of filaments and membranes in micrographs and tomograms, geometric instance separation of dense structures, and the alignment pipelines that make large volumes usable. The goal is throughput — analyses that took months of manual annotation running in minutes, reproducibly, at dataset scale.
Software
TARDIS-EM
Transformer-based rapid dimensionless instance segmentation. Combines deep semantic segmentation with a geometric model that separates individual objects, turning 2D and 3D micrographs — electron tomography, cryo-EM, fluorescence — into segmented point clouds. Pre-trained models for microtubules and membranes generalize across modalities and resolutions, and have been applied to over 13,000 tomograms from the CZI cryo-ET data portal.
GitHub · PyPI · napari hub · preprint
PANDORICA
Analytical tools for electron microscopy. Its serial-section tomogram stitcher solves the alignment problem in two stages: the image data fixes each section's global pose — rotation, shift, anisotropic stretch — while the microtubules drive only the fine residual warp, guarded by a diffeomorphism constraint so the deformation stays physically sane. Ships with a napari plugin for validating alignments on real datasets.
Publications
- Loading publications…