UPerNet
lastmod 2026-11-10created 2024-02-062 edits
UPerNet (Unified Perceptual Parsing Network) is a semantic segmentation model proposed for unified perceptual parsing, which recognizes visual concepts at multiple levels, such as scenes, objects, parts, materials, and textures, simultaneously with a single network. Its core is a decoder that combines a feature pyramid network (FPN) with a pyramid pooling module (PPM) to fuse features at multiple resolutions, and it has since been widely used as the standard decoder for evaluating the segmentation performance of backbones such as Swin Transformer and ConvNeXt.