Transformer-Based Attention Networks for Continuous Pixel-Wise Prediction

TL;DR

This paper introduces TransDepth, a novel architecture combining CNNs and transformers with a gated attention decoder, achieving state-of-the-art results in continuous pixel-wise prediction tasks like depth and surface normal estimation.

Contribution

It is the first to apply transformers to pixel-wise continuous label prediction, integrating a gated attention decoder to preserve local details.

Findings

Achieves state-of-the-art performance on three datasets

Effectively models long-range dependencies in pixel-wise tasks

Demonstrates the benefit of combining CNNs and transformers

Abstract

While convolutional neural networks have shown a tremendous impact on various computer vision tasks, they generally demonstrate limitations in explicitly modeling long-range dependencies due to the intrinsic locality of the convolution operation. Initially designed for natural language processing tasks, Transformers have emerged as alternative architectures with innate global self-attention mechanisms to capture long-range dependencies. In this paper, we propose TransDepth, an architecture that benefits from both convolutional neural networks and transformers. To avoid the network losing its ability to capture local-level details due to the adoption of transformers, we propose a novel decoder that employs attention mechanisms based on gates. Notably, this is the first paper that applies transformers to pixel-wise prediction problems involving continuous labels (i.e., monocular depth prediction and surface normal estimation). Extensive experiments demonstrate that the proposed TransDepth achieves state-of-the-art performance on three challenging datasets. Our code is available at: https://github.com/ygjwd12345/TransDepth.