A team of researchers from Twitter have published a paper detailing a machine learning technique that uses a generative adversarial network to make shrewd guesses about how to up-res small images by up to 400%, into crisp, large images, with eye-popping results.
Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function. Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios, but they are often lacking high-frequency details and are perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. In this paper, we present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4x upscaling factors. To achieve this, we propose a perceptual loss function which consists of an adversarial loss and a content loss. The adversarial loss pushes our solution to the natural image manifold using a discriminator network that is trained to differentiate between the super-resolved images and original photo-realistic images. In addition, we use a content loss motivated by perceptual similarity instead of similarity in pixel space. Our deep residual network is able to recover photo-realistic textures from heavily downsampled images on public benchmarks. An extensive mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are closer to those of the original high-resolution images than to those obtained with any state-of-the-art method.
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network [Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, Wenzhe Shi/Arxiv]
Photo-realistic single image super-resolution using a generative adversarial network [The Morning Paper]
(via Four Short Links)
The Econ-SF wiki is a new, annotated collaborative bibliography of science fiction that delves into economic topics -- remember that Paul Krugman was inspired to get into economics after reading Asimov's "Foundation" novels, to say nothing of all the people whose brains were colonized by Atlas Shrugged. It's brand new and has some notable omissions, […]
Our team of researchers at MIT’s Little Devices Lab have developed a pocket sized laboratory for biology that allows anyone to invent and deploy rapid diagnostics to detect diseases like Zika and Dengue, as well as everyday biomarkers like cholesterol. Using plug and play reaction blocks, it can be as easy as snapping Legos together. The current approach to developing diagnostic tools involves shipping out samples to faraway labs for the development of tests that take too long and cost too much - but what would happen if everyone could have the tools they needed to design and make diagnostics? If the ability to diagnose disease was directly in the hands of those who most needed it?
Canada's two leading digital rights groups, CIPPIC (previously) and Citizen Lab (previously) have issued a joint report called Shining a Light on the Encryption Debate: A Canadian Field Guide , and every Canadian should read it.
Handheld radios might seem a bit archaic, but in an emergency situation, few things will keep you as reliably connected to the outside world. This Emergency Multi-Function Radio & Flashlight takes the utility of the tried-and-true radio and combines it with a powerful flashlight and self-sufficient energy system. It’s available in the Boing Boing Store for […]
Few programming languages boast the versatility and user-friendliness of Python, which is why it’s the first language of choice for many aspiring programmers. Regardless of your experience level, you can take the first step to becoming Python-savvy with the Python 3 Bootcamp Bundle, available in the Boing Boing Store for $35 this week. Featuring more than […]
We live during a time where cyberattacks regularly make news headlines, so it should come as no surprise that cybersecurity professionals are experiencing a surge in demand at even the entry level, making now the ideal time to learn the tools of the trade if you’re considering a career switch. The 2018 Supercharged Cybersecurity Bundle offers […]