Large Kernel Size Research Articles

Efficient Photon Beam Diffusion for Directional Subsurface Scattering.

Real-time subsurface scattering techniques are widely used in translucent material rendering. Among advanced methods that rely on the bidirectional scattering-surface reflectance distribution function (BSSRDF), screen space algorithms exhibit limited translucency, while existing large-distance methods are inefficient and yield poor illumination details. To address these limitations for better large-distance scattering, we develop a novel algorithm by extending the photon beam diffusion (PBD) model within the light view and screen space. Unlike surface irradiance in prior methods, we incorporate the refracted beam in the medium into real-time scattering estimation, presenting a new consideration for photon beam utilization. Concretely, we store all photon beam samples in light view textures and utilize an adaptive sampling pattern for beam sample selection in large filtering kernel sizes. This can reduce the sample count based on surface attributes. In screen space, virtual sources are derived from samples to estimate PBD contributions, with an approximation that preserves boundary conditions. To avoid possible overestimation, we implement correction factors that scale contributions, effectively aligning our results with path-tracing references. Through these reformulations, our efficient PBD generates results closest to references among existing methods. The experiments accurately represent better front-face illumination details and backlit translucency effects, while significantly accelerating performance compared to previous large-distance methods.

Omni-Kernel Network for Image Restoration

Image restoration aims to reconstruct a high-quality image from a degraded low-quality observation. Recently, Transformer models have achieved promising performance on image restoration tasks due to their powerful ability to model long-range dependencies. However, the quadratically growing complexity with respect to the input size makes them inapplicable to practical applications. In this paper, we develop an efficient convolutional network for image restoration by enhancing multi-scale representation learning. To this end, we propose an omni-kernel module that consists of three branches, i.e., global, large, and local branches, to learn global-to-local feature representations efficiently. Specifically, the global branch achieves a global perceptive field via the dual-domain channel attention and frequency-gated mechanism. Furthermore, to provide multi-grained receptive fields, the large branch is formulated via different shapes of depth-wise convolutions with unusually large kernel sizes. Moreover, we complement local information using a point-wise depth-wise convolution. Finally, the proposed network, dubbed OKNet, is established by inserting the omni-kernel module into the bottleneck position for efficiency. Extensive experiments demonstrate that our network achieves state-of-the-art performance on 11 benchmark datasets for three representative image restoration tasks, including image dehazing, image desnowing, and image defocus deblurring. The code is available at https://github.com/c-yn/OKNet.

Impact of large kernel size on yield prediction: a case study of corn yield prediction with SEDLA in the U.S. Corn Belt

Predicting agricultural yields is imperative for effective planning to sustain the growing global population. Traditionally, regression-based, simulation-based, and hybrid methods were employed for yield prediction. In recent times, there has been a notable shift towards the adoption of Machine Learning (ML) methods, with Deep Learning (DL), particularly Convolutional Neural Networks (CNNs) and Long-Short Term Memory (LSTM) networks, emerging as popular choices for their enhanced predictive accuracy. This research introduces a cost-effective DL architecture tailored for corn yield prediction, considering computational efficiency in processing time, data size, and NN architecture complexity. The proposed architecture, named SEDLA (Simple and Efficient Deep Learning Architecture), leverages the spatial and temporal learning capabilities of CNNs and LSTMs, respectively, with a unique emphasis on exploring the impact of kernel size in CNNs. Simultaneously, the study aims to exclusively employ satellite and yield data, strategically minimizing input variables to enhance the model’s simplicity and efficiency. Notably, the study demonstrates that employing larger kernel sizes in CNNs, especially when processing histogram-based Surface Reflectance (SR) and Land Surface Temperature (LST) data from Moderate Resolution Imaging Spectroradiometer (MODIS), allows for a reduction in the number of hidden layers. The efficacy of the architecture was evaluated through extensive testing on corn yield prediction across 13 states in the United States (U.S.) Corn Belt at county-level. The experimental results showcase the superiority of the proposed architecture, achieving a Mean Absolute Percentage Error (MAPE) of 6.71 and Root Mean Square Error (RMSE) of 14.34, utilizing a single-layer CNN with a 15 × 15 kernel in conjunction with LSTM. These outcomes surpass existing benchmarks in the literature, affirming the efficacy and potential of the suggested DL framework for accurate and efficient crop yield predictions.

Hybrid optical convolutional neural network with convolution kernels trained in the spatial domain

Although many hybrid convolutional neural networks using optical convolution processors have been demonstrated for high energy efficiency and fast image processing speed, the need for accurate modeling of the optical convolution hardware and large Fourier kernel sizes result in a long and energy-intensive training process. Here, we demonstrate a hybrid convolutional neural network based on an optimized optical convolution processor—the system uses kernels trained in the spatial domain and compensates the optical path mismatch that arises from the reflection of digital micromirror devices for convolution computation. The spatial-domain convolution kernels are Fourier transformed and then binarized before being programmed into a digital micromirror device in the optical convolution processor. This method minimizes overhead from the optical convolution process in the training of the hybrid convolutional neural network, and it can improve the energy efficiency and reduce the training time of the hybrid optical convolutional neural network. For convolution computation using digital micromirror devices with grayscale images, bit-plane-wise convolution is also presented. We tested the hybrid neural network on MNIST, Fashion-MNIST, and CIFAR-10 color datasets and obtained classification accuracies of 98.7%, 85.8%, and 57.8%, respectively.

Multi-Scale Learning with Sparse Residual Network for Explainable Multi-Disease Diagnosis in OCT Images.

In recent decades, medical imaging techniques have revolutionized the field of disease diagnosis, enabling healthcare professionals to noninvasively observe the internal structures of the human body. Among these techniques, optical coherence tomography (OCT) has emerged as a powerful and versatile tool that allows high-resolution, non-invasive, and real-time imaging of biological tissues. Deep learning algorithms have been successfully employed to detect and classify various retinal diseases in OCT images, enabling early diagnosis and treatment planning. However, existing deep learning algorithms are primarily designed for single-disease diagnosis, which limits their practical application in clinical settings where OCT images often contain symptoms of multiple diseases. In this paper, we propose an effective approach for multi-disease diagnosis in OCT images using a multi-scale learning (MSL) method and a sparse residual network (SRN). Specifically, the MSL method extracts and fuses useful features from images of different sizes to enhance the discriminative capability of a classifier and make the disease predictions interpretable. The SRN is a minimal residual network, where convolutional layers with large kernel sizes are replaced with multiple convolutional layers that have smaller kernel sizes, thereby reducing model complexity while achieving a performance similar to that of existing convolutional neural networks. The proposed multi-scale sparse residual network significantly outperforms existing methods, exhibiting 97.40% accuracy, 95.38% sensitivity, and 98.25% specificity. Experimental results show the potential of our method to improve explainable diagnosis systems for various eye diseases via visual discrimination.

Monocular Road Scene Bird’s Eye View Prediction via Big Kernel-Size Encoder and Spatial-Channel Transform Module

A detailed representation of the surrounding road scene is crucial for an autonomous driving system. yellow The camera-based Bird’s Eye View map has been a popular solution to present the surrounding information, due to its low cost and rich spatial context information. Most of the existing methods predict the BEV map based on the depth-estimation or the trivial homography method, which may cause the error propagation and the absence of content. To overcome these drawbacks, we propose a novel end-to-end framework that employs the front monocular image to predict the road layout and vehicle occupancy. In particular, to capture the long-range feature, we redesign a CNN encoder with a large kernel size to extract the image features. For reducing the big difference between the front image features and the top-down features, we propose a novel Spatial-Channel projection module to convert the front map into the top-down space. Additionally, concerning the correlation between front view and top-down view, we propose the Dual Cross-view Transformer module to refine the top-down view feature maps and strengthen the transformation. Extensive evaluations on the KITTI and Argoverse datasets present that the proposed model achieves the state-of-the-art results for both datasets. Furthermore, the proposed model runs in 37 FPS on a single GPU, demonstrating the generation of a real-time BEV map. The code will be published at https://github.com/raozhongyu/BEV_LKA.

Deep-Learning-Based Low-Frequency Reconstruction in Full-Waveform Inversion

Low frequencies are vital for full-waveform inversion (FWI) to retrieve long-scale features and reliable subsurface properties from seismic data. Unfortunately, low frequencies are missing because of limitations in seismic acquisition steps. Furthermore, there is no explicit expression for transforming high frequencies into low frequencies. Therefore, low-frequency reconstruction (LFR) is imperative. Recently developed deep-learning (DL)-based LFR methods are based on either 1D or 2D convolutional neural networks (CNNs), which cannot take full advantage of the information contained in 3D prestack seismic data. Therefore, we present a DL-based LFR approach in which high frequencies are transformed into low frequencies by training an approximately symmetric encoding-decoding-type bridge-shaped 3D CNN. Our motivation is that the 3D CNN can naturally exploit more information that can be effectively used to improve the LFR result. We designed a Hanning-based window for suppressing the Gibbs effect associated with the hard splitting of the low- and high-frequency data. We report the significance of the convolutional kernel size on the training stage convergence rate and the performance of CNN’s generalization ability. CNN with reasonably large kernel sizes has a large receptive field and is beneficial to long-wavelength LFR. Experiments indicate that our approach can accurately reconstruct low frequencies from bandlimited high frequencies. The results of 3D CNN are distinctly superior to those of 2D CNN in terms of precision and highly relevant low-frequency energy. FWI on synthetic data indicates that the DL-predicted low frequencies nearly resemble those of actual low frequencies, and the DL-predicted low frequencies are accurate enough to mitigate the FWI’s cycle-skipping problems. Codes and data of this work are shared via a public repository.

A GCN-based fast CU partition method of intra-mode VVC

In this paper, a global convolutional network (GCN)-based fast coding unit (CU) partition method of intra-mode VVC is proposed. By using the GCN module with large kernel size convolutions, the proposed method can capture global information in CUs, leading to an accurate partition mode prediction in the quad-tree plus multi-type tree (QTMT) structure. Ranked according to predicted probabilities, the partition modes with lower probabilities are discarded, which reduces the computational complexity of VVC. Additionally, tradeoffs between performance and complexity can be achieved with different strategies. Experimental results demonstrated that the proposed method can reduce encoding time by 51.06%∼61.15% while increasing Bjøntegaard delta bit-rate (BD-BR) by 0.84%∼1.52% when implemented in VTM 10.0, outperforming the state-of-the-art methods, and that the proposed method can be used to accelerate VVenC 1.0 at the preset slower, achieving higher performance and lower complexity compared with the original VVenC 1.0 at the presets slow and medium.

Non-uniform Nyström approximation for sparse kernel regression: Theoretical analysis and experimental evaluation

Solving a kernel regression problem usually suffers from expensive computation and storage costs due to the large kernel size. To tackle this problem, the Nyström method is proposed and widely applied to large-scale kernel methods as an approximate solution. The key idea of this method is to select a subset of columns of the kernel matrix and rebuilds a low-rank approximation to the dense kernel matrix. To reduce computational costs of sparse kernel regression, we take the merits of the Nyström approximation and present two non-uniform Nyström methods with theoretical guarantees for sparse kernel regression in this paper. In detail, we first provide an upper bound to the solution of sparse kernel regression via Nyström approximation. Based on this bound, we prove the upper bounds of the optimal solutions when adopting two notable non-uniform landmark selection strategies, including Determinantal Point Processes (DPPs) and Ridge Leverage Scores (RLS). Compared with the uniform Nyström method, we empirically demonstrate the superior performance of non-uniform Nyström in sparse kernel regression on a synthetic dataset and several real-world datasets.

Strength properties of the Bambara kernel (Vigna subterranean) as influenced by the moisture content and kernel size

The fracture resistance of food grains is an essential piece of information required for the optimum design and development of agricultural post-harvest machinery. In this study, the strength properties of two varieties of Bambara kernels (TVSU-1395 and TVSU-1353) were examined in terms of the mean rupture force, absorbed energy, and deformation as affected by the moisture content and kernel size. To achieve this, a quasi-compressive force was applied on the two varieties of Bambara kernels of varying moisture contents (5.43%, 7.24%, 9.01%, 11.54%, and 13.62% wb) and kernel sizes (small, medium, and large) in between the loading compartments of a universal Testometric device at a 20 mm/min loading rate. The experiments take ten treatments with 20 replications subjected factorially to a completely randomised design (CRD) into consideration. The results revealed that the force needed to initiate the kernel fracture increased with an increase in the kernel size and moisture content from 101.44 to 235.06 N and 74.69 to 190.49 N for TVSU-1395 and TVSU-1353, respectively; whereas the energy at the kernel fracture point increased in a range of 0.074 to 0.401 J and 0.062 to 0.141 J for TVSU-1395 and TVSU-1353, respectively. The kernel deformation increased with the moisture content and size from 0.654 to 3.746 mm. These infer that the large kernel size of the TVSU-1395 variety at a 5.4% moisture content had greater compressive strength than the TVSU-1353 variety. The kernel moisture and size exhibited a strong correlation (0.958 ≤ R<sup>2</sup> ≤ 0.997) with the strength parameters. The results of this study will help the food industry in designing energy-efficient post-harvest equipment for Bambara kernel processing. Further studies may consider the strength attributes of Bambara kernels at varying rates of loading, kernel orientations, and varieties to optimise the best process conditions for the post-harvest handling of different Bambara cultivars and develop labour-saving decorticating machines.

Learnable filter-banks for CNN-based audio applications

We investigate the design of a convolutional layer where kernels are parameterized functions. This layer aims at being the input layer of convolutional neural networks for audio applications or applications involving time-series. The kernels are defined as one-dimensional functions having a band-pass filter shape, with a limited number of trainable parameters. Building on the literature on this topic, we confirm that networks having such an input layer can achieve state-of-the-art accuracy on several audio classification tasks. We explore the effect of different parameters on the network accuracy and learning ability. This approach reduces the number of weights to be trained and enables larger kernel sizes, an advantage for audio applications. Furthermore, the learned filters bring additional interpretability and a better understanding of the audio properties exploited by the network.

Salient Object Detection via Dynamic Scale Routing.

Recent research advances in salient object detection (SOD) could largely be attributed to ever-stronger multi-scale feature representation empowered by the deep learning technologies. The existing SOD deep models extract multi-scale features via the off-the-shelf encoders and combine them smartly via various delicate decoders. However, the kernel sizes in this commonly-used thread are usually "fixed". In our new experiments, we have observed that kernels of small size are preferable in scenarios containing tiny salient objects. In contrast, large kernel sizes could perform better for images with large salient objects. Inspired by this observation, we advocate the "dynamic" scale routing (as a brand-new idea) in this paper. It will result in a generic plug-in that could directly fit the existing feature backbone. This paper's key technical innovations are two-fold. First, instead of using the vanilla convolution with fixed kernel sizes for the encoder design, we propose the dynamic pyramid convolution (DPConv), which dynamically selects the best-suited kernel sizes w.r.t. the given input. Second, we provide a self-adaptive bidirectional decoder design to accommodate the DPConv-based encoder best. The most significant highlight is its capability of routing between feature scales and their dynamic collection, making the inference process scale-aware. As a result, this paper continues to enhance the current SOTA performance. Both the code and dataset are publicly available at https://github.com/wuzhenyubuaa/DPNet.

Novel encoder for ambient data compression applied to microcontrollers in agricultural robots

Agricultural robots are flexible to obtain ambient information across large areas of farmland. However, it needs to face two major challenges: data compression and filtering noise. To address these challenges, an encoder for ambient data compression, named Tiny-Encoder, was presented to compress and filter raw ambient information, which can be applied to agricultural robots. Tiny-Encoder is based on the operation of convolutions and pooling, and it has a small number of layers and filters. With the aim of evaluating the performance of Tiny-Encoder, different three types of ambient information (including temperature, humidity, and light) were selected to show the performance of compressing raw data and filtering noise. In the task of compressing raw data, Tiny-Encoder obtained higher accuracy (less than the maximum error of sensors ±0.5°C or ±3.5% RH) and more appropriate size (the largest size is 205 KB) than the other two auto-encoders based convolutional operations with different compressed features (including 20, 60, and 200 features). As for filtering noise, Tiny-Encoder has comparable performance with three conventional filtering approaches (including median filtering, Gaussian filtering, and Savitzky-Golay filtering). With large kernel size (i.e., 5), Tiny-Encoder has the best performance among these four filtering approaches: the coefficients of variation with the large kernel (i.e., 5) were 8.6189% (temperature), 10.2684% (humidity), 57.3576% (light), respectively. Overall, Tiny-Encoder can be used for ambient information compression applied to microcontrollers in agricultural information acquisition robots. Keywords: agricultural information, robot, ambient information, data compression, embedded machine learning methods DOI: 10.25165/j.ijabe.20221504.6911 Citation: Wang Y T, Li M Z, Ji R H, Wang M J, Zhang Y, Zheng L H. Novel encoder for ambient data compression applied to microcontrollers in agricultural robots. Int J Agric & Biol Eng, 2022; 15(4): 197–204.

Genetic analysis of the hybrid rice population ‘Mavr × Kontakt’

Rice can form not only white-color kernels, but also of red, brown and black colors. In black-color rice kernels, the pericarp contains anthocyanin, which has antioxidant activity and therefore has a positive effect on human health. The purpose of the current study was to develop rice samples with black pericarp. The paper has presented the study results of the hybrid population ‘Mavr × Kontakt’ of the second generation. The variety ‘Mavr’ has a black pericarp, the variety ‘Kontakt’ has a white one. In the process of hybridological analysis there have been identified the patterns of inheritance of the main quantitative traits affecting productivity; there have been identified the best samples, in which formed compact erect panicles and black kernels; there has been selected the initial material for breeding work. The study was carried out in 2020 on the plots of the ES “Proletarskoye” in the Rostov region. There was established that the color of pericarp was inherited according to the type of complementary interaction of two genes. There was found an overdominance and interaction of two pairs of genes of different strengths according to ‘plant height’. There was also seen partial positive dominance, transgressions, and digenic differences of the initial varieties according to ‘panicle length’. According to the traits ‘number of spikelets per panicle’, there was identified overdominance of large values and positive transgression. There were found the forms with well-kerneled panicles. The trait ‘1000 kernel weight’ was characterized by negative dominance and dihybrid cleavage of 9:6:1. There have been selected the best morphotype F2 forms with black pericarp, which possessed optimal plant height, long panicles, larger kernel size, and an average 1000 kernel weight.

Physical and hydration properties of specialty floury and sweet maize kernels subjected to pan and microwave toasting

The three main underutilized maize types from the Andean region have specialty floury and sweet kernels to make entire ready-to-eat maize by toasting in a pan until a golden brown surface color. The study aims to characterize these maize kernels before and after pan and microwave toasting. The initial moisture content for the raw kernel condition was the same (14.06 ± 0.13 g/100 g) to ensure data reliability by maize types and toasting conditions. The toasting conditions generated kernel expansions studied through the geometric, gravimetric, surface color, instrumental texture, milling average particle size, and hydration properties. Innate to maize types, the color properties (L*, a*, b*, C*, h*, and ΔE*) demonstrated that the toasted kernels developed a comparable range of surface color by the effect of pan and microwave toasting conditions (P ≥ 0.05). Large kernel sizes, low kernel hardness, and small average particle size ascribed to the endosperm softness were the main properties that characterize floury maize. High color differences, high water-soluble index, low density, high kernel hardness, and large average particle size confirmed the sugary-vitreous fraction in sweet maize. Microwave toasting produced lower remaining moisture content and kernel density, larger kernel size associated with the internal porosity created during toasting, higher water absorption capacity, and swelling power than pan toasting.

Convolutional fusion network for monaural speech enhancement

Convolutional neural network (CNN) based methods, such as the convolutional encoder–decoder network, offer state-of-the-art results in monaural speech enhancement. In the conventional encoder–decoder network, large kernel size is often used to enhance the model capacity, which, however, results in low parameter efficiency. This could be addressed by using group convolution, as in AlexNet, where group convolutions are performed in parallel in each layer, before their outputs are concatenated. However, with the simple concatenation, the inter-channel dependency information may be lost. To address this, the Shuffle network re-arranges the outputs of each group before concatenating them, by taking part of the whole input sequence as the input to each group of convolution. In this work, we propose a new convolutional fusion network (CFN) for monaural speech enhancement by improving model performance, inter-channel dependency, information reuse and parameter efficiency. First, a new group convolutional fusion unit (GCFU) consisting of the standard and depth-wise separable CNN is used to reconstruct the signal. Second, the whole input sequence (full information) is fed simultaneously to two convolution networks in parallel, and their outputs are re-arranged (shuffled) and then concatenated, in order to exploit the inter-channel dependency within the network. Third, the intra skip connection mechanism is used to connect different layers inside the encoder as well as decoder to further improve the model performance. Extensive experiments are performed to show the improved performance of the proposed method as compared with three recent baseline methods.

PFWNet: Pretraining neural network via feature jigsaw puzzle for weakly-supervised temporal action localization

Weakly supervised temporal action localization is a challenging yet interesting task. Existing methods usually apply a few temporal convolutional layers or linear layers to predict classification scores, where the model capacity is limited. Inspired by counterpart researches, increasing model capacity is the potential to improve the localization performance. However, under the weakly supervised paradigm, the video-level classification label is insufficient to learn large-capacity models. The essential reason lies in that most of the inputs to action localization networks are high-level features extracted by video recognition models. In lack of off-the-shelf initialization weights, the action localization networks have to train from scratch and can only explore low-capacity models. In this work, we are inspired by the self-supervised learning paradigm and propose to learn high-quality representative models via solving the feature jigsaw puzzle task. The proposed self-supervised pretraining process can explore networks with large kernel size and deeper layers, which can provide valuable initialization to action localization networks. In the implementation, we first discover potential action scopes via calculating motion intensity. Then, we cut features into snippets and permute them into out-of-order status. We randomly discard frames for boundaries between two snippets to guide the network learning high-level representations and prevent information leakage. Moreover, because the potential permutation number factorially rises with the increase of snippet number, we select a fixed number of permutation operations via the maximum hamming distance criterion, which eases the learning process. Comprehensive experiments on two benchmarks demonstrate the efficiency of pretraining to weakly supervised action localization task, and the proposed method builds new state-of-the-art performance.

Identification of North American softwoods via machine-learning

This manuscript reports the feasibility of a sequential convolutional neural network (CNN) machine-learning model that correctly identifies 11 North American softwood species from 14× magnified macroscopic end-grain images. The convolutional network contained a large kernel size, max pooling layers, and leaky rectified linear units to accelerate training. To reduce overfitting of training data, we employed L2 regularization, custom initialization, and stratified 5-fold cross-validation techniques. The database consisted of 1789 wood end-grain images. The training data set consisted of 1431 images, whereas the validation set had approximately 358 images. In both sets, the input image size was 227 pixels × 227 pixels. Data augmentation was performed on-the-fly by flipping, rotating, and zooming the images. We tested the performance of the CNN against precision, sensitivity, specificity, F1 score, and adjusted accuracy. The adjusted accuracy for the entire model was 94.0%. Confusion matrices indicated the lowest performance was in correctly classifying ponderosa pine (Pinus ponderosa Douglas ex P. Lawson & C. Lawson) and eastern spruce (Picea spp. A. Dietr.) group with an average sensitivity of 89.0% for each. Even though high validation accuracy (>94.0%) was achieved, we concluded that a much larger data set is needed for wood identification to obtain industrially accurate identification of softwoods, mainly due to their visual and macroscopic similarities.

SE-SqueezeNet: SqueezeNet extension with squeeze-and-excitation block

Convolutional neural networks have been popularly used for image recognition tasks. It is known that deep convolutional neural network can yield high recognition accuracy while training it can be very time-consuming. AlexNet was one of the very first networks shown to be effective for the tasks. However, due to its large kernel sizes and fully connected layers, the training time is significant. SqueezeNet has been known as smaller network that yields the same performance as AlexNet. Based on SqueezeNet, we are interested in exploring the effective insertion of the squeeze-and-excitation (SE) module into SqueezeNet that can further improve the performance and cost efficiency. The promising methodology and pattern of module insertion have been explored. The experimental results for evaluating the module insertion show the improvement on top1 accuracy by 1.55% and 3.32% while the model size is enlarged by up to 16% and 10% for CIFAR100 and ILSVRC2012 datasets respectively.

Context Module Based Multi-patch Hierarchical Network for Motion Deblurring

Single image blind motion deblurring refers to transferring a blurred motion image into a corresponding clear image, which is a challenging and classic problem in the field of computer vision. The spatially variant blur is usually caused by many factors, such as camera jitter and object motion. Since image deblurring can be regarded as the task of image transformation, deep learning methods based on coarse-to-fine scheme, especially those using multi-scale architectures become popular. However, they have the disadvantages of unsatisfactory image quality and time-consuming running caused by large kernel size. In this paper, we propose a novel end-to-end network structure based on Deep Hierarchical Multi-patch network architecture integrated with Context Module and additional ResBlocks in order to tackle deblurring problem. Compared with recently proposed networks, it generates images with better visual effect as well as higher image quality index. Besides, our model significantly reduces the test time. We evaluate the proposed network structure on public GoPro dataset, a large-scale image dataset with complex synthetic blur. The experiments on the benchmark dataset prove that our effective method outperforms other state-of-the-art blind deblurring algorithms both qualitatively and quantitatively, which demonstrates the effectiveness of Context Module in the task of single image blur removal.