In this no comment we focus on the growing technique of super-resolution. The idea is simple take a low resolution photo and scale it up with more pixels. The surprising thing is that it is possible to fill in more of the missing data than you might imagine using neural networks.
Sometimes the news is reported well enough elsewhere and we have little to add other than to bring it to your attention.
No Comment is a format where we present original source information, lightly edited, so that you can decide if you want to follow it up.
In this//No Comment we have an academic paper on implementing super-resolution using a neural network and a GitHub repro that gives you the chance to try out a neural network method in Python.
The usual way of increasing the number of pixels in an image, also called up-scaling is to use interpolation so that the new pixels have values that are intermediate between their surrounding pixels. The most common interpolation method is bi-cubic which is the method that sophisticated neural network methods should be compared to.
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?
During image downsampling information is lost, making super-resolution a highly ill-posed inverse problem with a large set of possible solutions. The behavior of optimization-based super-resolution methods is therefore principally driven by the choice of objective function. Recent work has largely focussed on minimizing the mean squared reconstruction error (MSE). The resulting estimates have high peak signal-to-noise-ratio (PSNR), but they are often overly smoothed, lack high-frequency detail, making them perceptually unsatisfying. In this paper, we present super-resolution generative adversarial network (SRGAN).
To our knowledge, it is the first framework capable of recovering photo-realistic natural images from 4 times downsampling. 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 function motivated by perceptual similarity instead of similarity in pixel space.
Trained on 350K images using the perceptual loss function, our deep residual network was able to recover photo-realistic textures from heavily downsampled images on public benchmarks.
Image super-resolution through deep learning. This project uses deep learning to upscale 16x16 images by a 4x factor. The resulting 64x64 images display sharp features that are plausible based on the dataset that was used to train the neural net.
Here's an random, non cherry-picked, example of what this network can do. From left to right, the first column is the 16x16 input image, the second one is what you would get from a standard bicubic interpolation, the third is the output generated by the neural net, and on the right is the ground truth.
This particular example was produced after training the network for 3 hours on a GTX 1080 GPU, equivalent to 130,000 batches or about 10 epochs.
To use the code you will need Python 3 with Tensorflow, numpy, scipy and moviepy.
This year's premium conference for PostgreSQL took place in Athens, Greece between October 22-25. The nice Athenian weather and cultural aspect aside, the conference was a big hit too.
Thomas Eugene Kurtz, the co-founder of the BASIC programming language, has died at the age of 96. BASIC, which was developed for the purpose of education, popularized computer programming making it ac [ ... ]
