PP-LiteSeg
lastmod 2026-11-10created 2025-03-113 edits
PP-LiteSeg is a real-time semantic segmentation model released by Baidu's PaddleSeg team. It consists of a Flexible and Lightweight Decoder that gradually reduces the number of decoder channels, a Unified Attention Fusion Module that fuses features with spatial and channel attention, and a Simple Pyramid Pooling Module that aggregates wide context with little computation, balancing accuracy and speed. On benchmarks such as Cityscapes it showed high throughput among lightweight models.
[2204.02681] PP-LiteSeg: A Superior Real-Time Semantic Segmentation Model
Documentation
- PP-LiteSeg - A Superior Real-Time Semantic Segmentation Model
- https://arxiv.org/abs/2204.02681