Vision Transformers for Dense Prediction
This work addresses the problem of achieving finer-grained and globally coherent predictions in computer vision tasks for researchers and practitioners, representing a novel method rather than an incremental improvement.
The paper tackled dense prediction tasks like depth estimation and semantic segmentation by introducing dense vision transformers as a backbone, replacing convolutional networks, and achieved improvements such as up to 28% better relative performance in depth estimation and a new state-of-the-art 49.02% mIoU on ADE20K for segmentation.
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.