Deep learning based superresolution achieves high-quality results, but its heavy computational workload, large buffer, and high external memory bandwidth inhibit its usage in mobile devices. To solve the above issues, this paper proposes a real-time hardware accelerator with the tilted layer fusion method that reduces the external DRAM bandwidth by 92\% and just needs 102KB on-chip memory. The design implemented with a 40nm CMOS process achieves 1920x1080@60fps throughput with 544.3K gate count when running at 600MHz; it has higher throughput and lower area cost than previous designs.
Real-world face super-resolution (SR) is a highly ill-posed image restoration task. The fully-cycled Cycle-GAN architecture is widely employed to achieve promising performance on face SR, but prone to produce artifacts upon challenging cases in real-world scenarios, since joint participation in the same degradation branch will impact final performance due to huge domain gap between real-world and synthetic LR ones obtained by generators. To better exploit the powerful generative capability of GAN for real-world face SR, in this paper, we establish two independent degradation branches in the forward and backward cycle-consistent reconstruction processes, respectively, while the two processes share the same restoration branch. Our Semi-Cycled Generative Adversarial Networks (SCGAN) is able to alleviate the adverse effects of the domain gap between the real-world LR face images and the synthetic LR ones, and to achieve accurate and robust face SR performance by the shared restoration branch regularized by both the forward and backward cycle-consistent learning processes. Experiments on two synthetic and two real-world datasets demonstrate that, our SCGAN outperforms the state-of-the-art methods on recovering the face structures/details and quantitative metrics for real-world face SR. The code will be publicly released at https://github.com/HaoHou-98/SCGAN.
Enormous efforts have been recently made to super-resolve hyperspectral (HS) images with the aid of high spatial resolution multispectral (MS) images. Most prior works usually perform the fusion task by means of multifarious pixel-level priors. Yet the intrinsic effects of a large distribution gap between HS-MS data due to differences in the spatial and spectral resolution are less investigated. The gap might be caused by unknown sensor-specific properties or highly-mixed spectral information within one pixel (due to low spatial resolution). To this end, we propose a subpixel-level HS super-resolution framework by devising a novel decoupled-and-coupled network, called DC-Net, to progressively fuse HS-MS information from the pixel- to subpixel-level, from the image- to feature-level. As the name suggests, DC-Net first decouples the input into common (or cross-sensor) and sensor-specific components to eliminate the gap between HS-MS images before further fusion, and then fully blends them by a model-guided coupled spectral unmixing (CSU) net. More significantly, we append a self-supervised learning module behind the CSU net by guaranteeing the material consistency to enhance the detailed appearances of the restored HS product. Extensive experimental results show the superiority of our method both visually and quantitatively and achieve a significant improvement in comparison with the state-of-the-arts. Furthermore, the codes and datasets will be available at https://sites.google.com/view/danfeng-hong for the sake of reproducibility.
The image Super-Resolution (SR) technique has greatly improved the visual quality of images by enhancing their resolutions. It also calls for an efficient SR Image Quality Assessment (SR-IQA) to evaluate those algorithms or their generated images. In this paper, we focus on the SR-IQA under deep learning and propose a Structure-and-Perception-based Quality Evaluation (SPQE). In emerging deep-learning-based SR, a generated high-quality, visually pleasing image may have different structures from its corresponding low-quality image. In such case, how to balance the quality scores between no-reference perceptual quality and referenced structural similarity is a critical issue. To help ease this problem, we give a theoretical analysis on this tradeoff and further calculate adaptive weights for the two types of quality scores. We also propose two deep-learning-based regressors to model the no-reference and referenced scores. By combining the quality scores and their weights, we propose a unified SPQE metric for SR-IQA. Experimental results demonstrate that the proposed method outperforms the state-of-the-arts in different datasets.
Due to the sophisticated imaging process, an identical scene captured by different cameras could exhibit distinct imaging patterns, introducing distinct proficiency among the super-resolution (SR) models trained on images from different devices. In this paper, we investigate a novel and practical task coded cross-device SR, which strives to adapt a real-world SR model trained on the paired images captured by one camera to low-resolution (LR) images captured by arbitrary target devices. The proposed task is highly challenging due to the absence of paired data from various imaging devices. To address this issue, we propose an unsupervised domain adaptation mechanism for real-world SR, named Dual ADversarial Adaptation (DADA), which only requires LR images in the target domain with available real paired data from a source camera. DADA employs the Domain-Invariant Attention (DIA) module to establish the basis of target model training even without HR supervision. Furthermore, the dual framework of DADA facilitates an Inter-domain Adversarial Adaptation (InterAA) in one branch for two LR input images from two domains, and an Intra-domain Adversarial Adaptation (IntraAA) in two branches for an LR input image. InterAA and IntraAA together improve the model transferability from the source domain to the target. We empirically conduct experiments under six Real to Real adaptation settings among three different cameras, and achieve superior performance compared with existing state-of-the-art approaches. We also evaluate the proposed DADA to address the adaptation to the video camera, which presents a promising research topic to promote the wide applications of real-world super-resolution. Our source code is publicly available at https://github.com/lonelyhope/DADA.git.
Transformer-based methods have shown impressive performance in low-level vision tasks, such as image super-resolution. However, we find that these networks can only utilize a limited spatial range of input information through attribution analysis. This implies that the potential of Transformer is still not fully exploited in existing networks. In order to activate more input pixels for better reconstruction, we propose a novel Hybrid Attention Transformer (HAT). It combines both channel attention and window-based self-attention schemes, thus making use of their complementary advantages of being able to utilize global statistics and strong local fitting capability. Moreover, to better aggregate the cross-window information, we introduce an overlapping cross-attention module to enhance the interaction between neighboring window features. In the training stage, we additionally adopt a same-task pre-training strategy to exploit the potential of the model for further improvement. Extensive experiments show the effectiveness of the proposed modules, and we further scale up the model to demonstrate that the performance of this task can be greatly improved. Our overall method significantly outperforms the state-of-the-art methods by more than 1dB. Codes and models are available at https://github.com/XPixelGroup/HAT.
Despite the plethora of successful Super-Resolution Reconstruction (SRR) models applied to natural images, their application to remote sensing imagery tends to produce poor results. Remote sensing imagery is often more complicated than natural images and has its peculiarities such as being of lower resolution, it contains noise, and often depicting large textured surfaces. As a result, applying non-specialized SRR models on remote sensing imagery results in artifacts and poor reconstructions. To address these problems, this paper proposes an architecture inspired by previous research work, introducing a novel approach for forcing an SRR model to output realistic remote sensing images: instead of relying on feature-space similarities as a perceptual loss, the model considers pixel-level information inferred from the normalized Digital Surface Model (nDSM) of the image. This strategy allows the application of better-informed updates during the training of the model which sources from a task (elevation map inference) that is closely related to remote sensing. Nonetheless, the nDSM auxiliary information is not required during production and thus the model infers a super-resolution image without any additional data besides its low-resolution pairs. We assess our model on two remotely sensed datasets of different spatial resolutions that also contain the DSM pairs of the images: the DFC2018 dataset and the dataset containing the national Lidar fly-by of Luxembourg. Based on visual inspection, the inferred super-resolution images exhibit particularly superior quality. In particular, the results for the high-resolution DFC2018 dataset are realistic and almost indistinguishable from the ground truth images.
Using mathematical modeling and human subjects experiments, this research explores the extent to which emerging webcams might leak recognizable textual and graphical information gleaming from eyeglass reflections captured by webcams. The primary goal of our work is to measure, compute, and predict the factors, limits, and thresholds of recognizability as webcam technology evolves in the future. Our work explores and characterizes the viable threat models based on optical attacks using multi-frame super resolution techniques on sequences of video frames. Our models and experimental results in a controlled lab setting show it is possible to reconstruct and recognize with over 75% accuracy on-screen texts that have heights as small as 10 mm with a 720p webcam. We further apply this threat model to web textual contents with varying attacker capabilities to find thresholds at which text becomes recognizable. Our user study with 20 participants suggests present-day 720p webcams are sufficient for adversaries to reconstruct textual content on big-font websites. Our models further show that the evolution towards 4K cameras will tip the threshold of text leakage to reconstruction of most header texts on popular websites. Besides textual targets, a case study on recognizing a closed-world dataset of Alexa top 100 websites with 720p webcams shows a maximum recognition accuracy of 94% with 10 participants even without using machine-learning models. Our research proposes near-term mitigations including a software prototype that users can use to blur the eyeglass areas of their video streams. For possible long-term defenses, we advocate an individual reflection testing procedure to assess threats under various settings, and justify the importance of following the principle of least privilege for privacy-sensitive scenarios.
Most of the existing video face super-resolution (VFSR) methods are trained and evaluated on VoxCeleb1, which is designed specifically for speaker identification and the frames in this dataset are of low quality. As a consequence, the VFSR models trained on this dataset can not output visual-pleasing results. In this paper, we develop an automatic and scalable pipeline to collect a high-quality video face dataset (VFHQ), which contains over $16,000$ high-fidelity clips of diverse interview scenarios. To verify the necessity of VFHQ, we further conduct experiments and demonstrate that VFSR models trained on our VFHQ dataset can generate results with sharper edges and finer textures than those trained on VoxCeleb1. In addition, we show that the temporal information plays a pivotal role in eliminating video consistency issues as well as further improving visual performance. Based on VFHQ, by analyzing the benchmarking study of several state-of-the-art algorithms under bicubic and blind settings. See our project page: https://liangbinxie.github.io/projects/vfhq
In medical imaging analysis, deep learning has shown promising results. We frequently rely on volumetric data to segment medical images, necessitating the use of 3D architectures, which are commended for their capacity to capture interslice context. However, because of the 3D convolutions, max pooling, up-convolutions, and other operations utilized in these networks, these architectures are often more inefficient in terms of time and computation than their 2D equivalents. Furthermore, there are few 3D pretrained model weights, and pretraining is often difficult. We present a simple yet effective 2D method to handle 3D data while efficiently embedding the 3D knowledge during training. We propose transforming volumetric data into 2D super images and segmenting with 2D networks to solve these challenges. Our method generates a super-resolution image by stitching slices side by side in the 3D image. We expect deep neural networks to capture and learn these properties spatially despite losing depth information. This work aims to present a novel perspective when dealing with volumetric data, and we test the hypothesis using CNN and ViT networks as well as self-supervised pretraining. While attaining equal, if not superior, results to 3D networks utilizing only 2D counterparts, the model complexity is reduced by around threefold. Because volumetric data is relatively scarce, we anticipate that our approach will entice more studies, particularly in medical imaging analysis.
Synthetic aperture radar (SAR) tomography (TomoSAR) has attracted remarkable interest for its ability in achieving three-dimensional reconstruction along the elevation direction from multiple observations. In recent years, compressed sensing (CS) technique has been introduced into TomoSAR considering for its super-resolution ability with limited samples. Whereas, the CS-based methods suffer from several drawbacks, including weak noise resistance, high computational complexity and complex parameter fine-tuning. Among the different CS algorithms, iterative soft-thresholding algorithm (ISTA) is widely used as a robust reconstruction approach, however, the parameters in the ISTA algorithm are manually chosen, which usually requires a time-consuming fine-tuning process to achieve the best performance. Aiming at efficient TomoSAR imaging, a novel sparse unfolding network named analytic learned ISTA (ALISTA) is proposed towards the TomoSAR imaging problem in this paper, and the key parameters of ISTA are learned from training data via deep learning to avoid complex parameter fine-tuning and significantly relieves the training burden. In addition, experiments verify that it is feasible to use traditional CS algorithms as training labels, which provides a tangible supervised training method to achieve better 3D reconstruction performance even in the absence of labeled data in real applications.
Decision making algorithms are used in a multitude of different applications. Conventional approaches for designing decision algorithms employ principled and simplified modelling, based on which one can determine decisions via tractable optimization. More recently, deep learning approaches that use highly parametric architectures tuned from data without relying on mathematical models, are becoming increasingly popular. Model-based optimization and data-centric deep learning are often considered to be distinct disciplines. Here, we characterize them as edges of a continuous spectrum varying in specificity and parameterization, and provide a tutorial-style presentation to the methodologies lying in the middle ground of this spectrum, referred to as model-based deep learning. We accompany our presentation with running examples in super-resolution and stochastic control, and show how they are expressed using the provided characterization and specialized in each of the detailed methodologies. The gains of combining model-based optimization and deep learning are demonstrated using experimental results in various applications, ranging from biomedical imaging to digital communications.
Modern Earth observation satellites capture multi-exposure bursts of push-frame images that can be super-resolved via computational means. In this work, we propose a super-resolution method for such multi-exposure sequences, a problem that has received very little attention in the literature. The proposed method can handle the signal-dependent noise in the inputs, process sequences of any length, and be robust to inaccuracies in the exposure times. Furthermore, it can be trained end-to-end with self-supervision, without requiring ground truth high resolution frames, which makes it especially suited to handle real data. Central to our method are three key contributions: i) a base-detail decomposition for handling errors in the exposure times, ii) a noise-level-aware feature encoding for improved fusion of frames with varying signal-to-noise ratio and iii) a permutation invariant fusion strategy by temporal pooling operators. We evaluate the proposed method on synthetic and real data and show that it outperforms by a significant margin existing single-exposure approaches that we adapted to the multi-exposure case.
Given a degraded input image, image restoration aims to recover the missing high-quality image content. Numerous applications demand effective image restoration, e.g., computational photography, surveillance, autonomous vehicles, and remote sensing. Significant advances in image restoration have been made in recent years, dominated by convolutional neural networks (CNNs). The widely-used CNN-based methods typically operate either on full-resolution or on progressively low-resolution representations. In the former case, spatial details are preserved but the contextual information cannot be precisely encoded. In the latter case, generated outputs are semantically reliable but spatially less accurate. This paper presents a new architecture with a holistic goal of maintaining spatially-precise high-resolution representations through the entire network, and receiving complementary contextual information from the low-resolution representations. The core of our approach is a multi-scale residual block containing the following key elements: (a) parallel multi-resolution convolution streams for extracting multi-scale features, (b) information exchange across the multi-resolution streams, (c) non-local attention mechanism for capturing contextual information, and (d) attention based multi-scale feature aggregation. Our approach learns an enriched set of features that combines contextual information from multiple scales, while simultaneously preserving the high-resolution spatial details. Extensive experiments on six real image benchmark datasets demonstrate that our method, named as MIRNet-v2 , achieves state-of-the-art results for a variety of image processing tasks, including defocus deblurring, image denoising, super-resolution, and image enhancement. The source code and pre-trained models are available at https://github.com/swz30/MIRNetv2
Convolutional neural network (CNN) based image enhancement methods such as super-resolution and detail enhancement have achieved remarkable performances. However, amounts of operations including convolution and parameters within the networks cost high computing power and need huge memory resource, which limits the applications with on-device requirements. Lightweight image enhancement network should restore details, texture, and structural information from low-resolution input images while keeping their fidelity. To address these issues, a lightweight image enhancement network is proposed. The proposed network include self-feature extraction module which produces modulation parameters from low-quality image itself, and provides them to modulate the features in the network. Also, dense modulation block is proposed for unit block of the proposed network, which uses dense connections of concatenated features applied in modulation layers. Experimental results demonstrate better performance over existing approaches in terms of both quantitative and qualitative evaluations.
Increasingly, convolution neural network (CNN) based super resolution models have been proposed for better reconstruction results, but their large model size and complicated structure inhibit their real-time hardware implementation. Current hardware designs are limited to a plain network and suffer from lower quality and high memory bandwidth requirements. This paper proposes a super resolution hardware accelerator with hardware efficient pixel attention that just needs 25.9K parameters and simple structure but achieves 0.38dB better reconstruction images than the widely used FSRCNN. The accelerator adopts full model block wise convolution for full model layer fusion to reduce external memory access to model input and output only. In addition, CNN and pixel attention are well supported by PE arrays with distributed weights. The final implementation can support full HD image reconstruction at 30 frames per second with TSMC 40nm CMOS process.
The development of the transformer-based text-to-image models are impeded by its slow generation and complexity for high-resolution images. In this work, we put forward a solution based on hierarchical transformers and local parallel auto-regressive generation. We pretrain a 6B-parameter transformer with a simple and flexible self-supervised task, Cross-modal general language model (CogLM), and finetune it for fast super-resolution. The new text-to-image system, CogView2, shows very competitive generation compared to concurrent state-of-the-art DALL-E-2, and naturally supports interactive text-guided editing on images.
In the last decade, the blossom of deep learning has witnessed the rapid development of scene text recognition. However, the recognition of low-resolution scene text images remains a challenge. Even though some super-resolution methods have been proposed to tackle this problem, they usually treat text images as general images while ignoring the fact that the visual quality of strokes (the atomic unit of text) plays an essential role for text recognition. According to Gestalt Psychology, humans are capable of composing parts of details into the most similar objects guided by prior knowledge. Likewise, when humans observe a low-resolution text image, they will inherently use partial stroke-level details to recover the appearance of holistic characters. Inspired by Gestalt Psychology, we put forward a Stroke-Aware Scene Text Image Super-Resolution method containing a Stroke-Focused Module (SFM) to concentrate on stroke-level internal structures of characters in text images. Specifically, we attempt to design rules for decomposing English characters and digits at stroke-level, then pre-train a text recognizer to provide stroke-level attention maps as positional clues with the purpose of controlling the consistency between the generated super-resolution image and high-resolution ground truth. The extensive experimental results validate that the proposed method can indeed generate more distinguishable images on TextZoom and manually constructed Chinese character dataset Degraded-IC13. Furthermore, since the proposed SFM is only used to provide stroke-level guidance when training, it will not bring any time overhead during the test phase. Code is available at https://github.com/FudanVI/FudanOCR/tree/main/text-gestalt.
Scene text image super-resolution (STISR) has been regarded as an important pre-processing task for text recognition from low-resolution scene text images. Most recent approaches use the recognizer's feedback as clues to guide super-resolution. However, directly using recognition clue has two problems: 1) Compatibility. It is in the form of probability distribution, has an obvious modal gap with STISR - a pixel-level task; 2) Inaccuracy. it usually contains wrong information, thus will mislead the main task and degrade super-resolution performance. In this paper, we present a novel method C3-STISR that jointly exploits the recognizer's feedback, visual and linguistical information as clues to guide super-resolution. Here, visual clue is from the images of texts predicted by the recognizer, which is informative and more compatible with the STISR task; while linguistical clue is generated by a pre-trained character-level language model, which is able to correct the predicted texts. We design effective extraction and fusion mechanisms for the triple cross-modal clues to generate a comprehensive and unified guidance for super-resolution. Extensive experiments on TextZoom show that C3-STISR outperforms the SOTA methods in fidelity and recognition performance. Code is available in https://github.com/zhaominyiz/C3-STISR.
Single-image super-resolution (SISR) has achieved significant breakthroughs with the development of deep learning. However, these methods are difficult to be applied in real-world scenarios since they are inevitably accompanied by the problems of computational and memory costs caused by the complex operations. To solve this issue, we propose a Lightweight Bimodal Network (LBNet) for SISR. Specifically, an effective Symmetric CNN is designed for local feature extraction and coarse image reconstruction. Meanwhile, we propose a Recursive Transformer to fully learn the long-term dependence of images thus the global information can be fully used to further refine texture details. Studies show that the hybrid of CNN and Transformer can build a more efficient model. Extensive experiments have proved that our LBNet achieves more prominent performance than other state-of-the-art methods with a relatively low computational cost and memory consumption. The code is available at https://github.com/IVIPLab/LBNet.