computer vision

稀疏补丁路由Transformer

Sparse Patch Routing Transformer

A neural network architecture that dynamically routes image patches to different computational depths based on local texture variance, class-token response, and boundary uncertainty to optimize efficiency in remote sensing scene segmentation. / 一种神经网络架构，根据局部纹理方差、类标记响应和边界不确定性动态将图像补丁路由到不同的计算深度，以优化遥感场景分割的效率。

Knowledge type: 工艺技术
First proposed or discovered: 2026
Proposer: Jihoon Kim, Minseo Park, Seungwoo Lee
Creator: 农场主
Last updated: 2026-11-05 18:06:49

Sub-knowledge points

SPRT (Sparse Patch Routing Transformer)saliency estimatorLoveDA datasetISPRS Vaihingen datasetWHU Building dataset

Higher-order knowledge points

Semantic Segmentation / 工艺技术remote sensing scene segmentationtransformer segmentation modelscomputational depth optimization

Main parameters and performance indicators

local texture variance: used for saliency estimationclass-token response: used for saliency estimationboundary uncertainty: used for saliency estimationshallow route: for low-texture background regionsstandard route: for common land-cover areasdetailed route: for boundary-sensitive regions

Open problems

No unresolved problem recorded yet

Related literature

ranked by community stars

b Sparse Patch Routing for Efficient Remote Sensing Scene Segmentation /b b /bOpen source

稀疏补丁路由Transformer

Sub-knowledge points

Higher-order knowledge points

Main parameters and performance indicators

Open problems

Related products and links