Browse Teach Curate

Shape-Guided Transformer Network for Remote Sensing Instance Segmentation

A transformer-based instance segmentation model for remote sensing images (SGTN) that pairs an LSwin encoder using vertical and horizontal 1D global self-attention for long-range context with a shape guidance module that emphasizes object boundaries, achieving accurate instance masks across single- and multi-class aerial datasets.

Data Text

2501.00360

This paper presents SGTN (Shape Guided Transformer Network), a deep-learning method for instance segmentation in remote sensing images. It introduces LSwin, a transformer encoder that applies separat…