Learning Recursive Filters for Low-Level Vision via a Hybrid Neural Network

By Sifei Liu, Jin-shan Pan, Ming-Hsuan Yang. 2016

168 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

Citations in the corpus

Semantic Image Inpainting with Deep Generative Models (2016, neutral): “Unlike previous hand-crafted matching and editing, learning based methods have shown promising results [27, 38, 33, 22].”
Learning Affinity via Spatial Propagation Networks (2017, methodology): “(9) has been used previously for semantic segmentation [4] and for learning low-level vision filters [19].”
Low-Light Image Enhancement via a Deep Hybrid Network (2019, methodology): “For this purpose, we introduce a novel spatially variant RNN inspired by [26].”
Dynamic Scene Deblurring Using Spatially Variant Recurrent Neural Networks (2018, methodology): “[19] develop a hybrid network including a CNN and RNN for image processing.”
Classification-Driven Dynamic Image Enhancement (2017, background): “Several recent works have shown that convolutional neural networks (CNN) [2, 3, 23, 27, 39, 40] can successfully WLS Filter [a]”
A Generic Deep Architecture for Single Image Reflection Removal and Image Smoothing (2017, methodology): “Table 5 shows that our method also significantly outperformed [24] for all the 4 filtering algorithms.”
Non-Local Spatial Propagation Network for Depth Completion (2020, methodology): “One way to reduce the bias is to limit the range of raw affinities [20], for example, to [−1/C, 1/C] using the hyperbolic tangent function (tanh(·)) with a normalization factor C.”
Deep bilateral learning for real-time image enhancement (2017, methodology): “earn an implicitly-defined operation from a database of human annotated input/output pairs. Many deep learning architectures have been used for image-toimage transformations, e.g. [Isola et al. 2016; Liu et al. 2016; Long et al. 2015; Xu et al. 2015; Yan et al. 2016]. However, most prior work incur ”
Seeing Motion in the Dark (2019, background): “Dense correspondence is almost impossible to be estimated accurately in extreme noisy conditions, which is required in some existing methods [27, 18, 15] but not in our system.”
Fast End-to-End Trainable Guided Filter (2018, background): “to accelerate a wide variety of image processing operators, which performed superior to the prior works [40, 31, 7, 26, 24], achieving the best results regarding speed and accuracy.”
Deep Structured Energy-Based Image Inpainting (2018, background): “Recently, deep learning has been employed for image inpainting and provided remarkable results [1], [10], [11].”
Edge-Enhanced GAN for Remote Sensing Image Superresolution (2019, methodology): “[48] used a hybrid CNN to approximate edge-preserving filters to extract and preserve the image edges.”
Deep Image Harmonization (2017, background): “Recently, neural network based methods for image editing tasks such as image colorization [7, 14, 30], inpainting [20] and filtering [18], have drawn much attention due to their efficiency and impressive results.”
Spatio-Temporal Filter Adaptive Network for Video Deblurring (2019, methodology): “However, compared with the video-based method, the accuracy of RNN weights is highly limited to having only a single blurry image as input.”
Fast Image Processing with Fully-Convolutional Networks (2017, background): “[51] combined a convolutional network and a set of recurrent networks to approximate a variety of image filters.”