Файл:Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network 1609.05158v2.pdf

Материал из Материалы по машинному обучению
Перейти к: навигация, поиск
Real-Time_Single_Image_and_Video_Super-Resolution_Using_an_Efficient_Sub-Pixel_Convolutional_Neural_Network_1609.05158v2.pdf(0 × 0 пикселей, размер файла: 3,3 МБ, MIME-тип: application/pdf)

Wenzhe Shi1, Jose Caballero1, Ferenc Husz´ar1, Johannes Totz1, Andrew P. Aitken1, Rob Bishop1, Daniel Rueckert1, Zehan Wang1 1Twitter 1{wshi,jcaballero,fhuszar,jtotz,aitken,rbishop,zehanw}@twitter.com

Abstract

Recently, several models based on deep neural networks haveachievedgreatsuccessintermsofbothreconstruction accuracy and computational performance for single image super-resolution. In these methods, the low resolution (LR) input image is upscaled to the high resolution (HR) space usingasinglefilter,commonlybicubicinterpolation,before reconstruction. This means that the super-resolution (SR) operation is performed in HR space. We demonstrate that this is sub-optimal and adds computational complexity. In thispaper,wepresentthefirstconvolutionalneuralnetwork (CNN) capable of real-time SR of 1080p videos on a single K2GPU.Toachievethis,weproposeanovelCNNarchitecture where the feature maps are extracted in the LR space. In addition, we introduce an efficient sub-pixel convolution layer which learns an array of upscaling filters to upscale the final LR feature maps into the HR output. By doing so, we effectively replace the handcrafted bicubic filter in the SR pipeline withmore complex upscaling filters specifically trained for each feature map, whilst also reducing the computational complexity of the overall SR operation. We evaluate the proposed approach using images and videos from publicly available datasets and show that it performs significantly better (+0.15dB on Images and +0.39dB on Videos) and is an order of magnitude faster than previous CNN-based methods.

История файла

Нажмите на дату/время, чтобы просмотреть, как тогда выглядел файл.

Дата/времяРазмерыУчастникПримечание
текущий18:38, 23 декабря 20160 × 0 (3,3 МБ)Slikos (обсуждение | вклад)
  • Вы не можете перезаписать этот файл.

Следующая 1 страница ссылается на данный файл: