---

λSplit: Self-Supervised Content-Aware Spectral Unmixing for Fluorescence Microscopy

Federico Carrara^1,2, Talley Lambert^1,3, Mehdi Seifi¹, Florian Jug¹

¹ Fondazione Human Technopole, Milan, Italy ² Università Campus Bio-Medico, Rome, Italy ³ Harvard Medical School, Boston, US

λSplit is a self-supervised method for spectral unmixing in fluorescence microscopy. It combines a Ladder VAE with a physics-based Spectral Mixer that encodes the image-formation model, enabling it to separate overlapping fluorophore emissions without ground-truth supervision. Compared to classical, pixel-wise unmixing methods, λSplit leverages spatial context, improving unmixing performance and robustness in challenging imaging regimes, such as in presence of considerable noise, highly overlapping spectra, or reduced spectral dimensionality.

A more detailed description of the method can be found in the preprint.

Presented at ECCV 2026.

Important

Code availability. The code for λSplit is not yet publicly available.

λSplit is the subject of pending patent applications held by Fondazione Human Technopole. We are finalizing the licensing terms for the public release, which we intend to make under a strong copyleft license, alongside a separate commercial licensing track for uses falling outside those terms.

We expect to release the code once these terms are in place. Until then, no license to the methods described in this repository or in the accompanying paper is granted, express or implied.

We are not accepting contributions at this time. A contributor license agreement will be in place before the code is released.

For licensing inquiries, contact federico.carrara@fht.org and florian.jug@fht.org.

To be notified when the code is released, click Watch above and select Custom → Releases.

Citation

If you find this work useful, please cite:

@inproceedings{carrara2026lambdasplit,
  title     = {λSplit: Self-Supervised Content-Aware Spectral Unmixing for Fluorescence Microscopy},
  author    = {Carrara, Federico and Lambert, Talley and Seifi, Mehdi and Jug, Florian},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}