iiinova — where your digital self keeps evolving

Susan Cox

@micoxscopy.bsky.social

Microscopy. Computer vision. Cell cytoskeleton. King's College London

Susan Cox @micoxscopy.bsky.social · 19/08/2025

Some fowl play here: vision language models are biased against 3-legged chickens, say they have 2 or 4 legs vlmsarebiased.github.io

A picture of a three legged chicken with the number of legs returned by different vision language models (2 or 4)

131

Susan Cox @micoxscopy.bsky.social · 12/08/2025

…and visualise structures and heterogeneity from the 8 nm alpha-tubulin repeat up to the mm scale.(3/5)

Top row shows spectrin rings in axons, middle row shows cell division, and bottom row shows plant trichomes.

130

Susan Cox @micoxscopy.bsky.social · 12/08/2025

By creating a structural model taking all the data into account and fitting the position, rotation, and heterogeneity in each observation of a structure, we can spot systematic biases in localisation microscopy data… (2/5)

Top row of image show nuclear pore structures derived from RESI data on the left and from 4Pi STORM and the right. Bottom graph shows that the RESI results show a correlation between the two rings of the nuclear pore structure and the z position, while the 4pi results show no such correlation

130