[2/2] For now, if you want per-channel measurements you must split your images by channel prior to running (not ideal if storage constrained). However, if you're interested in channel selection for N-channel TIFFs, or additional tooling for per-channel splitting, let me know and I can add support.

[1/2] There is currently only support for 1 and 3 channel images. No assumptions are made on the channel contents (e.g. RGB). For measurements, analytical descriptors are averaged over all channels and SSL features are computed on 3-channel images (1 channel are expanded if needed).

@alxndrkalinin.bsky.social As for python, I am writing bindings in a dedicated package github.com/tomouellette.... (mostly to satisfy the de facto numpy/zarr usage in python). Although the end goal is more of optimizations for large spatial analyses. It's WIP whenever I have extra time.

@alxndrkalinin.bsky.social `thyme` has ~100 core descriptors but computes up to ~300 if you choose a mode that measures foreground, background, etc. independently. On benchmarks, the predictive power is comparable to CellProfiler. `thyme` also has a few SSL models available via the CLI interface.

@Alex Kalinin If you also like the command-line, you may also like github.com/tomouellette.... A small CLI tool for going from images and/or masks to objects/features.