The state of the art in video super-resolution (SR) are techniques based on deep learning, but they perform poorly on real-world videos (see Figure 1). The reason is that training image-pairs are commonly created by downscaling a high-resolution image to produce a low-resolution counterpart. Deep models are therefore trained to undo downscaling and do not generalize to super-resolving real-world images. Several recent publications present techniques for improving the generalization of learning-based SR, but are all ill-suited for real-time application. We present a novel approach to synthesizing training data by simulating two digital-camera image-capture processes at different scales. Our method produces image-pairs in which both images have properties of natural images. Training an SR model using this data leads to far better generalization to real-world images and videos. In addition, deep video-SR models are characterized by a high operations-per-pixel count, which prohibits their application in real-time. We present an efficient CNN architecture, which enables real-time application of video SR on low-power edge-devices. We split the SR task into two sub-tasks: a control-flow which estimates global properties of the input video and adapts the weights and biases of a processing-CNN that performs the actual processing. Since the process-CNN is tailored to the statistics of the input, its capacity kept low, while retaining effectivity. Also, since video-statistics evolve slowly, the control-flow operates at a much lower rate than the video frame-rate. This reduces the overall computational load by as much as two orders of magnitude. This framework of decoupling the adaptivity of the algorithm from the pixel processing, can be applied in a large family of real-time video enhancement applications, e.g., video denoising, local tone-mapping, stabilization, etc.
Most video super-resolution methods focus on restoring high-resolution video frames from low-resolution videos without taking into account compression. However, most videos on the web or mobile devices are compressed, and the compression can be severe when the bandwidth is limited. In this paper, we propose a new compression-informed video super-resolution model to restore high-resolution content without introducing artifacts caused by compression. The proposed model consists of three modules for video super-resolution: bi-directional recurrent warping, detail-preserving flow estimation, and Laplacian enhancement. All these three modules are used to deal with compression properties such as the location of the intra-frames in the input and smoothness in the output frames. For thorough performance evaluation, we conducted extensive experiments on standard datasets with a wide range of compression rates, covering many real video use cases. We showed that our method not only recovers high-resolution content on uncompressed frames from the widely-used benchmark datasets, but also achieves state-of-the-art performance in super-resolving compressed videos based on numerous quantitative metrics. We also evaluated the proposed method by simulating streaming from YouTube to demonstrate its effectiveness and robustness. The source codes and trained models are available at https://github.com/google-research/google-research/tree/master/comisr.
In this work, we expand and test the capabilities of our recently developed super-resolution (SR) model to generate high-resolution (HR) realizations of the full phase-space matter distribution, including both displacement and velocity, from computationally cheap low-resolution (LR) cosmological N-body simulations. The SR model enhances the simulation resolution by generating 512 times more tracer particles, extending into the deeply non-linear regime where complex structure formation processes take place. We validate the SR model by deploying the model in 10 test simulations of box size 100 Mpc/h, and examine the matter power spectra, bispectra and 2D power spectra in redshift space. We find the generated SR field matches the true HR result at percent level down to scales of k ~ 10 h/Mpc. We also identify and inspect dark matter halos and their substructures. Our SR model generate visually authentic small-scale structures, that cannot be resolved by the LR input, and are in good statistical agreement with the real HR results. The SR model performs satisfactorily on the halo occupation distribution, halo correlations in both real and redshift space, and the pairwise velocity distribution, matching the HR results with comparable scatter, thus demonstrating its potential in making mock halo catalogs. The SR technique can be a powerful and promising tool for modelling small-scale galaxy formation physics in large cosmological volumes.
Brain image analysis has advanced substantially in recent years with the proliferation of neuroimaging datasets acquired at different resolutions. While research on brain image super-resolution has undergone a rapid development in the recent years, brain graph super-resolution is still poorly investigated because of the complex nature of non-Euclidean graph data. In this paper, we propose the first-ever deep graph super-resolution (GSR) framework that attempts to automatically generate high-resolution (HR) brain graphs with N' nodes (i.e., anatomical regions of interest (ROIs)) from low-resolution (LR) graphs with N nodes where N < N'. First, we formalize our GSR problem as a node feature embedding learning task. Once the HR nodes' embeddings are learned, the pairwise connectivity strength between brain ROIs can be derived through an aggregation rule based on a novel Graph U-Net architecture. While typically the Graph U-Net is a node-focused architecture where graph embedding depends mainly on node attributes, we propose a graph-focused architecture where the node feature embedding is based on the graph topology. Second, inspired by graph spectral theory, we break the symmetry of the U-Net architecture by super-resolving the low-resolution brain graph structure and node content with a GSR layer and two graph convolutional network layers to further learn the node embeddings in the HR graph. Third, to handle the domain shift between the ground-truth and the predicted HR brain graphs, we incorporate adversarial regularization to align their respective distributions. Our proposed AGSR-Net framework outperformed its variants for predicting high-resolution functional brain graphs from low-resolution ones. Our AGSR-Net code is available on GitHub at https://github.com/basiralab/AGSR-Net.
Diffusion-weighted MRI is nowadays performed routinely due to its prognostic ability, yet the quality of the scans are often unsatisfactory which can subsequently hamper the clinical utility. To overcome the limitations, here we propose a fully unsupervised quality enhancement scheme, which boosts the resolution and removes the motion artifact simultaneously. This process is done by first training the network using optimal transport driven cycleGAN with stochastic degradation block which learns to remove aliasing artifacts and enhance the resolution, then using the trained network in the test stage by utilizing bootstrap subsampling and aggregation for motion artifact suppression. We further show that we can control the trade-off between the amount of artifact correction and resolution by controlling the bootstrap subsampling ratio at the inference stage. To the best of our knowledge, the proposed method is the first to tackle super-resolution and motion artifact correction simultaneously in the context of MRI using unsupervised learning. We demonstrate the efficiency of our method by applying it to both quantitative evaluation using simulation study, and to in vivo diffusion-weighted MR scans, which shows that our method is superior to the current state-of-the-art methods. The proposed method is flexible in that it can be applied to various quality enhancement schemes in other types of MR scans, and also directly to the quality enhancement of apparent diffusion coefficient maps.
Single image super-resolution (SISR) aims to reconstruct high-resolution (HR) images from the given low-resolution (LR) ones, which is an ill-posed problem because one LR image corresponds to multiple HR images. Recently, learning-based SISR methods have greatly outperformed traditional ones, while suffering from over-smoothing, mode collapse or large model footprint issues for PSNR-oriented, GAN-driven and flow-based methods respectively. To solve these problems, we propose a novel single image super-resolution diffusion probabilistic model (SRDiff), which is the first diffusion-based model for SISR. SRDiff is optimized with a variant of the variational bound on the data likelihood and can provide diverse and realistic SR predictions by gradually transforming the Gaussian noise into a super-resolution (SR) image conditioned on an LR input through a Markov chain. In addition, we introduce residual prediction to the whole framework to speed up convergence. Our extensive experiments on facial and general benchmarks (CelebA and DIV2K datasets) show that 1) SRDiff can generate diverse SR results in rich details with state-of-the-art performance, given only one LR input; 2) SRDiff is easy to train with a small footprint; and 3) SRDiff can perform flexible image manipulation including latent space interpolation and content fusion.
Super-Resolution (SR) is a fundamental computer vision task that aims to obtain a high-resolution clean image from the given low-resolution counterpart. This paper reviews the NTIRE 2021 Challenge on Video Super-Resolution. We present evaluation results from two competition tracks as well as the proposed solutions. Track 1 aims to develop conventional video SR methods focusing on the restoration quality. Track 2 assumes a more challenging environment with lower frame rates, casting spatio-temporal SR problem. In each competition, 247 and 223 participants have registered, respectively. During the final testing phase, 14 teams competed in each track to achieve state-of-the-art performance on video SR tasks.
A recurrent structure is a popular framework choice for the task of video super-resolution. The state-of-the-art method BasicVSR adopts bidirectional propagation with feature alignment to effectively exploit information from the entire input video. In this study, we redesign BasicVSR by proposing second-order grid propagation and flow-guided deformable alignment. We show that by empowering the recurrent framework with the enhanced propagation and alignment, one can exploit spatiotemporal information across misaligned video frames more effectively. The new components lead to an improved performance under a similar computational constraint. In particular, our model BasicVSR++ surpasses BasicVSR by 0.82 dB in PSNR with similar number of parameters. In addition to video super-resolution, BasicVSR++ generalizes well to other video restoration tasks such as compressed video enhancement. In NTIRE 2021, BasicVSR++ obtains three champions and one runner-up in the Video Super-Resolution and Compressed Video Enhancement Challenges. Codes and models will be released to MMEditing.
Machine learning models are typically made available to potential client users via inference APIs. Model extraction attacks occur when a malicious client uses information gleaned from queries to the inference API of a victim model $F_V$ to build a surrogate model $F_A$ with comparable functionality. Recent research has shown successful model extraction of image classification, and natural language processing models. In this paper, we show the first model extraction attack against real-world generative adversarial network (GAN) image translation models. We present a framework for conducting such attacks, and show that an adversary can successfully extract functional surrogate models by querying $F_V$ using data from the same domain as the training data for $F_V$. The adversary need not know $F_V$'s architecture or any other information about it beyond its intended task. We evaluate the effectiveness of our attacks using three different instances of two popular categories of image translation: (1) Selfie-to-Anime and (2) Monet-to-Photo (image style transfer), and (3) Super-Resolution (super resolution). Using standard performance metrics for GANs, we show that our attacks are effective. Furthermore, we conducted a large scale (125 participants) user study on Selfie-to-Anime and Monet-to-Photo to show that human perception of the images produced by $F_V$ and $F_A$ can be considered equivalent, within an equivalence bound of Cohen's d = 0.3. Finally, we show that existing defenses against model extraction attacks (watermarking, adversarial examples, poisoning) do not extend to image translation models.
In this study, we propose a tailored DL framework for patient-specific performance that leverages the behavior of a model intentionally overfitted to a patient-specific training dataset augmented from the prior information available in an ART workflow - an approach we term Intentional Deep Overfit Learning (IDOL). Implementing the IDOL framework in any task in radiotherapy consists of two training stages: 1) training a generalized model with a diverse training dataset of N patients, just as in the conventional DL approach, and 2) intentionally overfitting this general model to a small training dataset-specific the patient of interest (N+1) generated through perturbations and augmentations of the available task- and patient-specific prior information to establish a personalized IDOL model. The IDOL framework itself is task-agnostic and is thus widely applicable to many components of the ART workflow, three of which we use as a proof of concept here: the auto-contouring task on re-planning CTs for traditional ART, the MRI super-resolution (SR) task for MRI-guided ART, and the synthetic CT (sCT) reconstruction task for MRI-only ART. In the re-planning CT auto-contouring task, the accuracy measured by the Dice similarity coefficient improves from 0.847 with the general model to 0.935 by adopting the IDOL model. In the case of MRI SR, the mean absolute error (MAE) is improved by 40% using the IDOL framework over the conventional model. Finally, in the sCT reconstruction task, the MAE is reduced from 68 to 22 HU by utilizing the IDOL framework.
Deep CNNs have achieved significant successes in image processing and its applications, including single image super-resolution (SR). However, conventional methods still resort to some predetermined integer scaling factors, e.g., x2 or x4. Thus, they are difficult to be applied when arbitrary target resolutions are required. Recent approaches extend the scope to real-valued upsampling factors, even with varying aspect ratios to handle the limitation. In this paper, we propose the SRWarp framework to further generalize the SR tasks toward an arbitrary image transformation. We interpret the traditional image warping task, specifically when the input is enlarged, as a spatially-varying SR problem. We also propose several novel formulations, including the adaptive warping layer and multiscale blending, to reconstruct visually favorable results in the transformation process. Compared with previous methods, we do not constrain the SR model on a regular grid but allow numerous possible deformations for flexible and diverse image editing. Extensive experiments and ablation studies justify the necessity and demonstrate the advantage of the proposed SRWarp method under various transformations.
Single Image Super-resolution (SISR) produces high-resolution images with fine spatial resolutions from aremotely sensed image with low spatial resolution. Recently, deep learning and generative adversarial networks(GANs) have made breakthroughs for the challenging task of single image super-resolution (SISR). However, thegenerated image still suffers from undesirable artifacts such as, the absence of texture-feature representationand high-frequency information. We propose a frequency domain-based spatio-temporal remote sensingsingle image super-resolution technique to reconstruct the HR image combined with generative adversarialnetworks (GANs) on various frequency bands (TWIST-GAN). We have introduced a new method incorporatingWavelet Transform (WT) characteristics and transferred generative adversarial network. The LR image hasbeen split into various frequency bands by using the WT, whereas, the transfer generative adversarial networkpredicts high-frequency components via a proposed architecture. Finally, the inverse transfer of waveletsproduces a reconstructed image with super-resolution. The model is first trained on an external DIV2 Kdataset and validated with the UC Merceed Landsat remote sensing dataset and Set14 with each image sizeof 256x256. Following that, transferred GANs are used to process spatio-temporal remote sensing images inorder to minimize computation cost differences and improve texture information. The findings are comparedqualitatively and qualitatively with the current state-of-art approaches. In addition, we saved about 43% of theGPU memory during training and accelerated the execution of our simplified version by eliminating batchnormalization layers.
Space-time video super-resolution (STVSR) aims to increase the spatial and temporal resolutions of low-resolution and low-frame-rate videos. Recently, deformable convolution based methods have achieved promising STVSR performance, but they could only infer the intermediate frame pre-defined in the training stage. Besides, these methods undervalued the short-term motion cues among adjacent frames. In this paper, we propose a Temporal Modulation Network (TMNet) to interpolate arbitrary intermediate frame(s) with accurate high-resolution reconstruction. Specifically, we propose a Temporal Modulation Block (TMB) to modulate deformable convolution kernels for controllable feature interpolation. To well exploit the temporal information, we propose a Locally-temporal Feature Comparison (LFC) module, along with the Bi-directional Deformable ConvLSTM, to extract short-term and long-term motion cues in videos. Experiments on three benchmark datasets demonstrate that our TMNet outperforms previous STVSR methods. The code is available at https://github.com/CS-GangXu/TMNet.
This paper presents deep unfolding neural networks to handle inverse problems in photothermal radiometry enabling super resolution (SR) imaging. Photothermal imaging is a well-known technique in active thermography for nondestructive inspection of defects in materials such as metals or composites. A grand challenge of active thermography is to overcome the spatial resolution limitation imposed by heat diffusion in order to accurately resolve each defect. The photothermal SR approach enables to extract high-frequency spatial components based on the deconvolution with the thermal point spread function. However, stable deconvolution can only be achieved by using the sparse structure of defect patterns, which often requires tedious, hand-crafted tuning of hyperparameters and results in computationally intensive algorithms. On this account, Photothermal-SR-Net is proposed in this paper, which performs deconvolution by deep unfolding considering the underlying physics. This enables to super resolve 2D thermal images for nondestructive testing with a substantially improved convergence rate. Since defects appear sparsely in materials, Photothermal-SR-Net applies trained block-sparsity thresholding to the acquired thermal images in each convolutional layer. The performance of the proposed approach is evaluated and discussed using various deep unfolding and thresholding approaches applied to 2D thermal images. Subsequently, studies are conducted on how to increase the reconstruction quality and the computational performance of Photothermal-SR-Net is evaluated. Thereby, it was found that the computing time for creating high-resolution images could be significantly reduced without decreasing the reconstruction quality by using pixel binning as a preprocessing step.
Recently, deep convolutional neural networks (CNNs) have been widely explored in single image super-resolution (SISR) and contribute remarkable progress. However, most of the existing CNNs-based SISR methods do not adequately explore contextual information in the feature extraction stage and pay little attention to the final high-resolution (HR) image reconstruction step, hence hindering the desired SR performance. To address the above two issues, in this paper, we propose a two-stage attentive network (TSAN) for accurate SISR in a coarse-to-fine manner. Specifically, we design a novel multi-context attentive block (MCAB) to make the network focus on more informative contextual features. Moreover, we present an essential refined attention block (RAB) which could explore useful cues in HR space for reconstructing fine-detailed HR image. Extensive evaluations on four benchmark datasets demonstrate the efficacy of our proposed TSAN in terms of quantitative metrics and visual effects. Code is available at https://github.com/Jee-King/TSAN.
In the era of artificial intelligence, convolutional neural networks (CNNs) are emerging as a powerful technique for computational imaging. They have shown superior quality for reconstructing fine textures from badly-distorted images and have potential to bring next-generation cameras and displays to our daily life. However, CNNs demand intensive computing power for generating high-resolution videos and defy conventional sparsity techniques when rendering dense details. Therefore, finding new possibilities in regular sparsity is crucial to enable large-scale deployment of CNN-based computational imaging. In this paper, we consider a fundamental but yet well-explored approach -- algebraic sparsity -- for energy-efficient CNN acceleration. We propose to build CNN models based on ring algebra that defines multiplication, addition, and non-linearity for n-tuples properly. Then the essential sparsity will immediately follow, e.g. n-times reduction for the number of real-valued weights. We define and unify several variants of ring algebras into a modeling framework, RingCNN, and make comparisons in terms of image quality and hardware complexity. On top of that, we further devise a novel ring algebra which minimizes complexity with component-wise product and achieves the best quality using directional ReLU. Finally, we implement an accelerator, eRingCNN, in two settings, n=2 and 4 (50% and 75% sparsity), with 40 nm technology to support advanced denoising and super-resolution at up to 4K UHD 30 fps. Layout results show that they can deliver equivalent 41 TOPS using 3.76 W and 2.22 W, respectively. Compared to the real-valued counterpart, our ring convolution engines for n=2 achieve 2.00x energy efficiency and 2.08x area efficiency with similar or even better image quality. With n=4, the efficiency gains of energy and area are further increased to 3.84x and 3.77x with 0.11 dB drop of PSNR.
The latest video coding standard, called versatile video coding (VVC), includes several novel and refined coding tools at different levels of the coding chain. These tools bring significant coding gains with respect to the previous standard, high efficiency video coding (HEVC). However, the encoder may still introduce visible coding artifacts, mainly caused by coding decisions applied to adjust the bitrate to the available bandwidth. Hence, pre and post-processing techniques are generally added to the coding pipeline to improve the quality of the decoded video. These methods have recently shown outstanding results compared to traditional approaches, thanks to the recent advances in deep learning. Generally, multiple neural networks are trained independently to perform different tasks, thus omitting to benefit from the redundancy that exists between the models. In this paper, we investigate a learning-based solution as a post-processing step to enhance the decoded VVC video quality. Our method relies on multitask learning to perform both quality enhancement and super-resolution using a single shared network optimized for multiple degradation levels. The proposed solution enables a good performance in both mitigating coding artifacts and super-resolution with fewer network parameters compared to traditional specialized architectures.
Convolutional neural networks have allowed remarkable advances in single image super-resolution (SISR) over the last decade. Among recent advances in SISR, attention mechanisms are crucial for high-performance SR models. However, the attention mechanism remains unclear on why and how it works in SISR. In this work, we attempt to quantify and visualize attention mechanisms in SISR and show that not all attention modules are equally beneficial. We then propose attention in attention network (A$^2$N) for more efficient and accurate SISR. Specifically, A$^2$N consists of a non-attention branch and a coupling attention branch. A dynamic attention module is proposed to generate weights for these two branches to suppress unwanted attention adjustments dynamically, where the weights change adaptively according to the input features. This allows attention modules to specialize to beneficial examples without otherwise penalties and thus greatly improve the capacity of the attention network with few parameters overhead. Experimental results demonstrate that our final model A$^2$N could achieve superior trade-off performances comparing with state-of-the-art networks of similar sizes. Codes are available at https://github.com/haoyuc/A2N.
The recent progress of deep convolutional neural networks has enabled great success in single image super-resolution (SISR) and many other vision tasks. Their performances are also being increased by deepening the networks and developing more sophisticated network structures. However, finding an optimal structure for the given problem is a difficult task, even for human experts. For this reason, neural architecture search (NAS) methods have been introduced, which automate the procedure of constructing the structures. In this paper, we expand the NAS to the super-resolution domain and find a lightweight densely connected network named DeCoNASNet. We use a hierarchical search strategy to find the best connection with local and global features. In this process, we define a complexity-based penalty for solving image super-resolution, which can be considered a multi-objective problem. Experiments show that our DeCoNASNet outperforms the state-of-the-art lightweight super-resolution networks designed by handcraft methods and existing NAS-based design.
Current deep image super-resolution (SR) approaches aim to restore high-resolution images from down-sampled images or by assuming degradation from simple Gaussian kernels and additive noises. However, these techniques only assume crude approximations of the real-world image degradation process, which should involve complex kernels and noise patterns that are difficult to model using simple assumptions. In this paper, we propose a more realistic process to synthesise low-resolution images for real-world image SR by introducing a new Kernel Adversarial Learning Super-resolution (KASR) framework. In the proposed framework, degradation kernels and noises are adaptively modelled rather than explicitly specified. Moreover, we also propose a high-frequency selective objective and an iterative supervision process to further boost the model SR reconstruction accuracy. Extensive experiments validate the effectiveness of the proposed framework on real-world datasets.