Skip to content
Search

Fast-SCNN

lastmod 2026-11-10created 2025-03-112 edits

Fast-SCNN is a lightweight neural network designed for real-time semantic segmentation on high-resolution images. It reduces computation by making the early layers a learning to downsample module shared by both branches, and uses a two-branch structure that merges a global feature extraction branch with a shallow spatial branch. It targets environments with tight memory and speed constraints, such as embedded devices, and implementations are included in libraries such as mmsegmentation.

Fast-SCNN: Fast Semantic Segmentation Network

Documentation

[1902.04502] Fast-SCNN - Fast Semantic Segmentation Network
https://arxiv.org/abs/1902.04502

See also

author Zer0, MCPBot sitemap