Vision Transformers for Dense Prediction

TL;DR

This paper introduces dense vision transformers that replace convolutional backbones for dense prediction tasks, achieving superior accuracy and global coherence, especially with large datasets.

Contribution

It proposes a novel dense vision transformer architecture that maintains high-resolution processing and global receptive fields, improving dense prediction performance over traditional convolutional networks.

Findings

Up to 28% improvement in monocular depth estimation.

New state-of-the-art results on ADE20K semantic segmentation.

Effective fine-tuning on smaller datasets like NYUv2 and KITTI.

Abstract

We introduce dense vision transformers, an architecture that leverages vision transformers in place of convolutional networks as a backbone for dense prediction tasks. We assemble tokens from various stages of the vision transformer into image-like representations at various resolutions and progressively combine them into full-resolution predictions using a convolutional decoder. The transformer backbone processes representations at a constant and relatively high resolution and has a global receptive field at every stage. These properties allow the dense vision transformer to provide finer-grained and more globally coherent predictions when compared to fully-convolutional networks. Our experiments show that this architecture yields substantial improvements on dense prediction tasks, especially when a large amount of training data is available. For monocular depth estimation, we observe an improvement of up to 28% in relative performance when compared to a state-of-the-art fully-convolutional network. When applied to semantic segmentation, dense vision transformers set a new state of the art on ADE20K with 49.02% mIoU. We further show that the architecture can be fine-tuned on smaller datasets such as NYUv2, KITTI, and Pascal Context where it also sets the new state of the art. Our models are available at https://github.com/intel-isl/DPT.