Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers

TL;DR

This paper introduces a novel transformer-based approach for semantic segmentation, treating it as a sequence-to-sequence task, which outperforms traditional convolutional methods on several benchmarks.

Contribution

Proposes SETR, a pure transformer model for semantic segmentation, offering an alternative to encoder-decoder architectures and achieving state-of-the-art results.

Findings

SETR achieves 50.28% mIoU on ADE20K

SETR attains 55.83% mIoU on Pascal Context

First place on ADE20K test leaderboard at submission time

Abstract

Most recent semantic segmentation methods adopt a fully-convolutional network (FCN) with an encoder-decoder architecture. The encoder progressively reduces the spatial resolution and learns more abstract/semantic visual concepts with larger receptive fields. Since context modeling is critical for segmentation, the latest efforts have been focused on increasing the receptive field, through either dilated/atrous convolutions or inserting attention modules. However, the encoder-decoder based FCN architecture remains unchanged. In this paper, we aim to provide an alternative perspective by treating semantic segmentation as a sequence-to-sequence prediction task. Specifically, we deploy a pure transformer (ie, without convolution and resolution reduction) to encode an image as a sequence of patches. With the global context modeled in every layer of the transformer, this encoder can be combined with a simple decoder to provide a powerful segmentation model, termed SEgmentation TRansformer (SETR). Extensive experiments show that SETR achieves new state of the art on ADE20K (50.28% mIoU), Pascal Context (55.83% mIoU) and competitive results on Cityscapes. Particularly, we achieve the first position in the highly competitive ADE20K test server leaderboard on the day of submission.