Fast Image Processing with Fully-Convolutional Networks

  title={Fast Image Processing with Fully-Convolutional Networks},
  author={Qifeng Chen and Jia Xu and Vladlen Koltun},
  journal={2017 IEEE International Conference on Computer Vision (ICCV)},
  • Qifeng Chen, Jia Xu, V. Koltun
  • Published 2 September 2017
  • Computer Science
  • 2017 IEEE International Conference on Computer Vision (ICCV)
We present an approach to accelerating a wide variety of image processing operators. [] Key Result We show that our models general- ize across datasets and across resolutions, and investigate a number of extensions of the presented approach.

Figures and Tables from this paper

Learned perceptual image enhancement
This paper shows that adding a learned no-reference image quality metric to the loss can significantly improve enhancement operators and can be effective for tuning a variety of operators such as local tone mapping and dehazing.
Restorable Image Operators with Quasi-Invertible Networks
A quasi-invertible model that learns common image processing operators in a restorable fashion is proposed that can generate visually pleasing results with the original content embedded and can be easily applied to practical applications such as restorable human face retouching and highlight preserved exposure adjustment.
Evaluating Parameterization Methods for Convolutional Neural Network (CNN)-Based Image Operators
Analysis of the operation principles of parameterization techniques and performance comparisons between image operators parameterized by using these methods are assessed experimentally on common image processing tasks including image smoothing, denoising, deblocking, and super-resolution.
Image smoothing via unsupervised learning
A unified unsupervised (label-free) learning framework that facilitates generating flexible and high-quality smoothing effects by directly learning from data using deep convolutional neural networks (CNNs).
CNN Based Image Restoration
An artificial neural network model based on deep neural networks is proposed to restore images damaged by inadequate sensor exposure, saturation, and underexposure, at the time of acquisition, which is adequate, considering the variability in equipment and photography techniques.
Region-Adaptive Dense Network for Efficient Motion Deblurring
This paper presents 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, enabling almost real-time deblurring.
An empirical evaluation of convolutional neural networks for image enhancement
This work empirically evaluate common architectures and loss functions employed for automatic image enhancement, and proposes an effective architecture-agnostic method for integrating additional contextual information into the enhancement process.
Decouple Learning for Parameterized Image Operators
A new decouple learning algorithm to learn from the operator parameters to dynamically adjust the weights of a deep network for image operators, denoted as the base network is proposed.
Spatially Variant Linear Representation Models for Joint Filtering
An effective algorithm based on a deep convolutional neural network (CNN) is developed that is able to estimate the spatially variant linear representation coefficients which are able to model the structural information of both the guidance and input images.
Fully Convolutional Network with Multi-Step Reinforcement Learning for Image Processing
This paper extends deep RL to pixelRL, and proposes an effective learning method for pixelRL that significantly improves the performance by considering not only the future states of the own pixel but also those of the neighbor pixels.


Natural Image Denoising with Convolutional Networks
An approach to low-level vision is presented that combines the use of convolutional networks as an image processing architecture and an unsupervised learning procedure that synthesizes training samples from specific noise models to avoid computational difficulties in MRF approaches that arise from probabilistic learning and inference.
Perceptual Losses for Real-Time Style Transfer and Super-Resolution
This work considers image transformation problems, and proposes the use of perceptual loss functions for training feed-forward networks for image transformation tasks, and shows results on image style transfer, where aFeed-forward network is trained to solve the optimization problem proposed by Gatys et al. in real-time.
Very Deep Convolutional Networks for Large-Scale Image Recognition
This work investigates the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting using an architecture with very small convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers.
Image Super-Resolution Using Deep Convolutional Networks
We propose a deep learning method for single image super-resolution (SR). Our method directly learns an end-to-end mapping between the low/high-resolution images. The mapping is represented as a deep
On learning optimized reaction diffusion processes for effective image restoration
This work extends conventional nonlinear reaction diffusion models by several parametrized linear filters as well as several parametsrized influence functions, and proposes to train the parameters of the filters and the influence functions through a loss based approach.
Bilateral guided upsampling
An algorithm to accelerate a large class of image processing operators by fitting local curves that map the input to the output that faithfully models state-of-the-art operators for tone mapping, style transfer, and recoloring is presented.
Accurate Image Super-Resolution Using Very Deep Convolutional Networks
This work presents a highly accurate single-image superresolution (SR) method using a very deep convolutional network inspired by VGG-net used for ImageNet classification and uses extremely high learning rates enabled by adjustable gradient clipping.
Deep Convolutional Neural Network for Image Deconvolution
This work develops a deep convolutional neural network to capture the characteristics of degradation, establishing the connection between traditional optimization-based schemes and a neural network architecture where a novel, separable structure is introduced as a reliable support for robust deconvolution against artifacts.
U-Net: Convolutional Networks for Biomedical Image Segmentation
It is shown that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks.
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
SRGAN, a generative adversarial network (GAN) for image super-resolution (SR), is presented, to its knowledge, the first framework capable of inferring photo-realistic natural images for 4x upscaling factors and a perceptual loss function which consists of an adversarial loss and a content loss.