HIMEC: Directional Change Representation and Fixed-Interface Decoding for Remote Sensing Image Change Captioning

Aysha Ashra, Author 2, Author 3, Author 4, Author 5

multimodalimage captioningremote sensingdeep learning

2608.12502

Builder Relevance

80%

6h ago

Abstract

HIMEC introduces a novel approach to remote sensing image change captioning by utilizing Directional Change Representation and fixed-interface decoding to improve semantic change description.

Reality Card

Core Claim

HIMEC achieves a CIDEr score of 142.81 on LEVIR-CC, outperforming traditional methods that use direct fused-feature memory.

Method / Result

Achieved a mean cosine distance of 0.69 on changed LEVIR-CC validation pairs.

Limitations

Findings are limited to the evaluated cascade, which may not generalize to other architectures.

← Back to all papers