Training Transformer Models by Wavelet Losses Improves Quantitative and Visual Performance in Single Image Super-Resolution

TL;DR

This paper introduces wavelet losses and convolutional non-local sparse attention blocks to enhance Transformer-based image super-resolution, achieving state-of-the-art quantitative and visual results.

Contribution

It presents a novel combination of wavelet losses with an extended hybrid Transformer architecture for improved super-resolution performance.

Findings

Achieves state-of-the-art PSNR results.

Provides superior visual quality in super-resolution tasks.

Demonstrates effectiveness of wavelet losses in training Transformer models.

Abstract

Transformer-based models have achieved remarkable results in low-level vision tasks including image super-resolution (SR). However, early Transformer-based approaches that rely on self-attention within non-overlapping windows encounter challenges in acquiring global information. To activate more input pixels globally, hybrid attention models have been proposed. Moreover, training by solely minimizing pixel-wise RGB losses, such as L1, have been found inadequate for capturing essential high-frequency details. This paper presents two contributions: i) We introduce convolutional non-local sparse attention (NLSA) blocks to extend the hybrid transformer architecture in order to further enhance its receptive field. ii) We employ wavelet losses to train Transformer models to improve quantitative and subjective performance. While wavelet losses have been explored previously, showing their power in training Transformer-based SR models is novel. Our experimental results demonstrate that the proposed model provides state-of-the-art PSNR results as well as superior visual performance across various benchmark datasets.