scverse/spatialdata: v0.7.3a0

LucaMarconato; Giovanni Palla; Wouter-Michiel Vierdag; Kevin Yamauchi; Isaac Virshup; Quentin Blampey; aeisenbarth; Arne Defauw; Tim Treis; Lukas Heumos; Benjamin Rombaut; Mark Keller; jonas2612; Tong LI; Sricharan Reddy Varra; Sonja-Stockhaus; Pariksheet Nanda; Nick ODell; Mikaela Koutrouli; Max Hargreaves; Marvin Albert; Martin Helm; LLehner; Ilan Gold; Grzegorz Bokota; Gregor Sturm; Florian Wuennemann; Elyas Heidari; Dylan Lam

doi:10.5281/zenodo.18611728

Published February 11, 2026 | Version v0.7.3a0

Software Open

scverse/spatialdata: v0.7.3a0

1. European Molecular Biology Laboratory
2. Iber Lab - ETH Zürich
3. CZI
4. Cure51 and Gustave Roussy Institute
5. @scverse @laminlabs @theislab
6. BioLizard
7. Wellcome Sanger Institute
8. @czbiohub-sf @mehta-lab
9. Genentech
10. ETH Zurich
11. Helmholtz Zentrum
12. University of Warsaw, Faculty of Mathematics, Informatics, and Mechanics
13. Boehringer Ingelheim
14. Seqera

What's Changed

Changed

New lazy to_multiscale() for labels by @LucaMarconato in https://github.com/scverse/spatialdata/pull/1068

Details: to_multiscale() was calling compute() for labels, leading to poor performance. This release changes the to_multiscale() default implementation to a lazy one (using resize() from scikit-image). Pixels for smaller scales are expected to have some numeric differences with respect to the old implementation (dask_image_nearest via multiscale-spatial-image==2.0.3). The old behavior can be replicated by passing method='dask_image_nearest' to the parsers for raster types.

Full Changelog: https://github.com/scverse/spatialdata/compare/v0.7.2...v0.7.3a0

Files