Search

Deep learning-based approaches have demonstrated impressive performance in single-image super-resolution (SISR). Efficient super-resolution compromises the reconstructed image’s quality to have fewer parameters and Flops. Ensured efficiency in image reconstruction and improved reconstruction quality of the model are significant challenges. This paper proposes a trio branch module (TBM) based on structural reparameterization. TBM achieves equivalence transformation through structural reparameterization operations, which use a complex network structure in the training phase and convert it to a more lightweight structure in the inference, achieving efficient inference while maintaining accuracy. Based on the TBM, we further design a lightweight version of the enhanced spatial attention mini (ESA-mini) and the residual trio feature block (RTFB). Moreover, the multiple RTFBs are combined to construct the residual trio network (RTFN). Finally, we introduce a localized contrast loss for better applicability to the super-resolution task, which enhances the reconstruction quality of the super-resolution model. Experiments show that the RTFN framework proposed in this paper outperforms other state-of-the-art efficient super-resolution methods in terms of inference speed and reconstruction quality. [ABSTRACT FROM AUTHOR]

Abstract Deep learning-based approaches have demonstrated impressive performance in single-image super-resolution (SISR). Efficient super-resolution compromises the reconstructed image’s quality to have fewer parameters and Flops. Ensured efficiency in image reconstruction and improved reconstruction quality of the model are significant challenges. This paper proposes a trio branch module (TBM) based on structural reparameterization. TBM achieves equivalence transformation through structural reparameterization operations, which use a complex network structure in the training phase and convert it to a more lightweight structure in the inference, achieving efficient inference while maintaining accuracy. Based on the TBM, we further design a lightweight version of the enhanced spatial attention mini (ESA-mini) and the residual trio feature block (RTFB). Moreover, the multiple RTFBs are combined to construct the residual trio network (RTFN). Finally, we introduce a localized contrast loss for better applicability to the super-resolution task, which enhances the reconstruction quality of the super-resolution model. Experiments show that the RTFN framework proposed in this paper outperforms other state-of-the-art efficient super-resolution methods in terms of inference speed and reconstruction quality.

In this paper, we aim to improve the imaging quality of super-resolution (SR) without increasing the inference time to address the difficulty of trading off between quality and inference time in many existing methods and design a deployment-friendly, lightweight model for mobile devices. Specifically, we propose a general RepDFSR framework to enhance the textures of SR images while avoiding additional inference time overhead, which can be applied to existing SR networks. It incorporates innovative convolutional block design, loss function design, and the re-parameterizable technique. In RepDFSR, we propose a re-parameterizable texture-enhanced convolution based on differential filters. It extracts texture information more advantageous and efficient in training than regular convolution. Secondly, we propose a DF loss function to compel the model to super-resolve the gradient mappings with high variance, thus reconstructing images with sharper textures. Moreover, we propose a TELNet network for mobile devices based on the RepDFSR framework to validate the effectiveness of RepDFSR and our thoughts on practical applications of SR on mobile devices. The experimental results demonstrate the successful integration of the RepDFSR framework with the existing SR methods. Additionally, the TELNet meets the requirements of mobile devices for hardware and quantization limitations, showcasing superior SR performance compared to classic and state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Single image super-resolution is a well-established low-level vision task that aims to reconstruct high-resolution images from low-resolution images. Methods based on Transformer have shown remarkable success and achieved outstanding performance in SISR tasks. While Transformer effectively models global information, it is less effective at capturing high frequencies such as stripes that primarily provide local information. Additionally, it has the potential to further enhance the capture of global information. To tackle this, we propose a novel Large Kernel Hybrid Attention Transformer using re-parameterization. It combines different kernel sizes and different steps re-parameterized convolution layers with Transformer to effectively capture global and local information to learn comprehensive features with low-frequency and high-frequency information. Moreover, in order to solve the problem of using batch normalization layer to introduce artifacts in SISR, we propose a new training strategy which is fusing convolution layer and batch normalization layer after certain training epochs. This strategy can enjoy the acceleration convergence effect of batch normalization layer in training and effectively eliminate the problem of artifacts in the inference stage. For re-parameterization of multiple parallel branch convolution layers, adopting this strategy can further reduce the amount of calculation of training. By coupling these core improvements, our LKHAT achieves state-of-the-art performance for single image super-resolution task. [Display omitted] • The proposed module takes into account both high-frequency and low-frequency information of the image • The proposed module is conducive to generating clear texture details • The training strategy we proposed helps to reduce artifacts [ABSTRACT FROM AUTHOR]