Researchers from graphics company NVIDIA developed a deep learning system that automatically reconstructs corrupted images or fills in missing parts:
Read the rest
The method, which performs a process called “image inpainting”, could be implemented in photo editing software to remove unwanted content, while filling it with a realistic computer-generated alternative.
“Our model can robustly handle holes of any shape, size location, or distance from the image borders. Previous deep learning approaches have focused on rectangular regions located around the center of the image, and often rely on expensive post-processing,” the NVIDIA researchers stated in their research paper. “Further, our model gracefully handles holes of increasing size.”
British cops got software to help them identify children in abuse images, but it's having trouble telling the difference between skin and sand.
...the department is working with “Silicon Valley providers” to help train the AI to successfully scan for images of child abuse. But as we’ve seen, even the most powerful tech companies can’t seem to deploy algorithms that don’t fuck up every once in awhile. They have promoted dangerous misinformation, abetted racism, and accidentally censored bisexual content. And when Gizmodo recently tested an app intended to automatically identify explicit images in your camera roll, it flagged a photo of a dog, a donut, and a fully-clothed Grace Kelly.
What are the odds that Londons' finest paid fabulous sums of money for what amounts to a dumb imagemagick script that averages and compares color samples from various points on each image, and are now paying fabulous sums more to have some rinkydink machine learning grafted on. Read the rest
The original is a screengrab of a fairly low-quality video feed; I opened it in a proprietary "Blow Up" app, added some grain to conceal compression artifacts, and interpolated it to 2048 pixels wide to get a better look at what president-elect Donald Trump was angry about. He looks quite charming, if you ask me! Now, promise not to use this image anywhere else, as it would be unseemly and unmannerly.
You can follow Rob on Twitter
— Lance Burson (@lanceburson) November 24, 2016
At the Rio olympics, every inch of the tennis arena beyond the court itself is bright lime green. Therefore, factionman chroma keyed it against a variety of interesting and charming backdrops. Read the rest
The HiRise imager in orbit around Mars shoots a continuous stream of data about its surface our way. Nasa's posted 44,000 images so far, each available in all sorts of formats and projections. You could have one a day as your desktop background and never run out.
Olivia Jack's Pixel Synth turns images into music, scanning across the pixels horizontally and interpreting brightness values as notes. The results are peculiar, obviously, but also strangely melodic. You can edit your image, too, or simply start with a blank canvas. Click "invert" for a synthesized moment of Hammer Horror. Advanced trippers can also edit the synth parameters and the drawing tool. Read the rest
The British Library has posted over a million copyright free images taken from books prior to 1900 on Flickr. That means if you need decorations of virtually any type for a website or book, you’ll find more than you can imagine among these visual riches. Just click through!