Region-Adaptive Dense Network for Efficient Motion Deblurring

Abstract

In this paper, we address the problem of dynamic scene deblurring in the presence of motion blur. Restoration of images affected by severe blur necessitates a network design with a large receptive field, which existing networks attempt to achieve through simple increment in the number of generic convolution layers, kernel-size, or the scales at which the image is processed. However, these techniques ignore the non-uniform nature of blur, and they come at the expense of an increase in model size and inference time. We present a new architecture composed of region adaptive dense deformable modules that implicitly discover the spatially varying shifts responsible for non-uniform blur in the input image and learn to modulate the filters. This capability is complemented by a self-attentive module which captures non-local spatial relationships among the intermediate features and enhances the spatially varying processing capability. We incorporate these modules into a densely connected encoder-decoder design which utilizes pre-trained Densenet filters to further improve the performance. Our network facilitates interpretable modeling of the spatially-varying deblurring process while dispensing with multi-scale processing and large filters entirely. Extensive comparisons with prior art on benchmark dynamic scene deblurring datasets clearly demonstrate the superiority of the proposed networks via significant improvements in accuracy and speed, enabling almost real-time deblurring.

Improving Audio Anomalies Recognition Using Temporal Convolutional Attention Networks

“…The architecture of the used self-attention unit [17] is shown in Figure 1(c). The input features are transformed into G and H via 1D convolution, and then generate the attention weights W from G and H by…”

Multi-Frame Blind Restoration for Image of Space Target With FRC and Branch-Attention

“…Liu et al [82] merged self-attention mechanism with RNN, and improved the robustness of network by utilizing RNN's characteristics of information broadcast. Purohit et al [41] proposed a motion deblur network with fast execution speed and strong performance ability by combining deformable convolution and self-attention mechanism. In this paper, we used a modified self-attention mechanism as that used in [54] and based on it we designed a branch-attention mechanism to broadcast information between all the restoration branches.…”

“…Many popular restoration methods based on CNNs tend to give preference to a U-net structure [40]. A normal part in this structure is the encoder-decoder module [41], [42]. Firstly, the encoder performs downsamplings on input features through pooling operation, and then the decoder does the corresponding up-sample operations to restore features.…”

“…At the beginning, the attention mechanisms that CNNs based on are channel attention [51] and spatial attention [52]. After Wang et al [53] proposing to use non-local relation of pixels between image features, the self-attention has been used widely in image restoration [41], [54] till now. In this paper, we proposed a branch-attention mechanism which based on self-attention to share effective information from branch to branch in the network.…”

LSRN: A Recurrent Residual Learning Framework for Continuous Wireless Channel Estimation Using Super-Resolution Concept

“…Feature extraction has been proved as a key procedure in the image analysis tasks, including SR, de-blurring [35], and image semantic segmentation [36]. With the modern machine learning technology, the intrinsic features are automatically learned by feeding the ''labelled'' data-sets, and the residual learning approach, such as residual network (ResNet), is reported to achieve the state-of-the-art performance for feature extraction in the image recognition tasks [37].…”