STAND: Semantic Anchoring Constraint with Dual-Granularity Disambiguation for Remote Sensing Image Change Captioning

The paper introduces STAND, a method for remote sensing image change captioning that tackles ambiguities in viewpoint, scale, and prior knowledge. It employs a semantic anchoring constraint to regularize temporal representations, a dual‑granularity disambiguation module that uses global context and frequency‑refocused attention to resolve spatial uncertainties, and a semantic concept anchoring module that leverages language priors during decoding. Experiments demonstrate that STAND outperforms existing approaches and effectively addresses these ambiguities.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.