AI RESEARCH
SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentation
arXiv CS.CV
•
ArXi:2605.22658v1 Announce Type: new While large language models provide strong compositional reasoning, existing reasoning segmentation pipelines fail to transparently connect this reasoning to visual perception. Current methods, such as latent query alignment, are end-to-end yet opaque "black boxes". Conversely, textual localization readout is merely readable, not truly interpretable, often functioning as an unconstrained post-hoc step.