DRCT: Saving Image Super-resolution away from Information Bottleneck

31 Mar 2024  ·  Chih-Chung Hsu, Chia-Ming Lee, Yi-Shiuan Chou ·

In recent years, Vision Transformer-based approaches for low-level vision tasks have achieved widespread success. Unlike CNN-based models, Transformers are more adept at capturing long-range dependencies, enabling the reconstruction of images utilizing non-local information. In the domain of super-resolution, Swin-transformer-based models have become mainstream due to their capability of global spatial information modeling and their shifting-window attention mechanism that facilitates the interchange of information between different windows. Many researchers have enhanced model performance by expanding the receptive fields or designing meticulous networks, yielding commendable results. However, we observed that it is a general phenomenon for the feature map intensity to be abruptly suppressed to small values towards the network's end. This implies an information bottleneck and a diminishment of spatial information, implicitly limiting the model's potential. To address this, we propose the Dense-residual-connected Transformer (DRCT), aimed at mitigating the loss of spatial information and stabilizing the information flow through dense-residual connections between layers, thereby unleashing the model's potential and saving the model away from information bottleneck. Experiment results indicate that our approach surpasses state-of-the-art methods on benchmark datasets and performs commendably at the NTIRE-2024 Image Super-Resolution (x4) Challenge. Our source code is available at https://github.com/ming053l/DRCT

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Super-Resolution BSD100 - 4x upscaling DRCT-L PSNR 28.16 # 1
SSIM 0.7557 # 5
Image Super-Resolution Manga109 - 4x upscaling DRCT-L PSNR 33.14 # 1
SSIM 0.9347 # 1
Image Super-Resolution Set14 - 4x upscaling DRCT-L PSNR 29.54 # 1
SSIM 0.8025 # 4
Image Super-Resolution Set5 - 4x upscaling DRCT-L PSNR 33.37 # 1
SSIM 0.9090 # 1
Image Super-Resolution Urban100 - 4x upscaling DRCT-L PSNR 28.70 # 1
SSIM 0.8508 # 2