Imaging The Past With Time-Travel Rephotography

April 12, 2021 by Adam Zeloof 49 Comments

Have you ever noticed that people in old photographs looks a bit weird? Deep wrinkles, sunken cheeks, and exaggerated blemishes are commonplace in photos taken up to the early 20th century. Surely not everybody looked like this, right? Maybe it was an odd makeup trend — was it just a fashionable look back then?

Not quite — it turns out that the culprit here is the film itself. The earliest glass-plate emulsions used in photography were only sensitive to the highest-frequency light, that which fell in the blue to ultraviolet range. Perhaps unsurprisingly, when combined with the fact that humans have red blood, this posed a real problem. While some of the historical figures we see in old photos may have benefited from an improved skincare regimen, the primary source of their haunting visage was that the photographic techniques available at the time were simply incapable of capturing skin properly. This lead to the sharp creases and dark lips we’re so used to seeing.

Of course, primitive film isn’t the only thing separating antique photos from the 42 megapixel behemoths that your camera can take nowadays. Film processing steps had the potential to introduce dust and other blemishes to the image, and over time the prints can fade and age in a variety of ways that depend upon the chemicals they were processed in. When rolled together, all of these factors make it difficult to paint an accurate portrait of some of history’s famous faces. Before you start to worry that you’ll never know just what Abraham Lincoln looked like, you might consider taking a stab at Time-Travel Rephotography.

Amazingly, Time-Travel Rephotography is a technique that actually lives up to how cool its name is. It uses a neural network (specifically, the StyleGAN2 framework) to take an old photo and project it into the space of high-res modern photos the network was trained on. This allows it to perform colorization, skin correction, upscaling, and various noise reduction and filtering operations in a single step which outputs remarkable results. Make sure you check out the project’s website to see some of the outputs at full-resolution.

We’ve seen AI upscaling before, but this project takes it to the next level by completely restoring antique photographs. We’re left wondering what techniques will be available 100 years from now to restore JPEGs stored way back in 2021, bringing them up to “modern” viewing standards.

Thanks to [Gus] for the tip!

Continue reading “Imaging The Past With Time-Travel Rephotography” →

Death Of The Turing Test In An Age Of Successful AIs

April 6, 2021 by Steven Dufresne 98 Comments

IBM has come up with an automatic debating system called Project Debater that researches a topic, presents an argument, listens to a human rebuttal and formulates its own rebuttal. But does it pass the Turing test? Or does the Turing test matter anymore?

The Turing test was first introduced in 1950, often cited as year-one for AI research. It asks, “Can machines think?”. Today we’re more interested in machines that can intelligently make restaurant recommendations, drive our car along the tedious highway to and from work, or identify the surprising looking flower we just stumbled upon. These all fit the definition of AI as a machine that can perform a task normally requiring the intelligence of a human. Though as you’ll see below, Turing’s test wasn’t even for intelligence or even for thinking, but rather to determine a test subject’s sex.

Continue reading “Death Of The Turing Test In An Age Of Successful AIs” →

AI Upscaling And The Future Of Content Delivery

April 5, 2021 by Tom Nardi 29 Comments

The rumor mill has recently been buzzing about Nintendo’s plans to introduce a new version of their extremely popular Switch console in time for the holidays. A faster CPU, more RAM, and an improved OLED display are all pretty much a given, as you’d expect for a mid-generation refresh. Those upgraded specifications will almost certainly come with an inflated price tag as well, but given the incredible demand for the current Switch, a $50 or even $100 bump is unlikely to dissuade many prospective buyers.

But according to a report from Bloomberg, the new Switch might have a bit more going on under the hood than you’d expect from the technologically conservative Nintendo. Their sources claim the new system will utilize an NVIDIA chipset capable of Deep Learning Super Sampling (DLSS), a feature which is currently only available on high-end GeForce RTX 20 and GeForce RTX 30 series GPUs. The technology, which has already been employed by several notable PC games over the last few years, uses machine learning to upscale rendered images in real-time. So rather than tasking the GPU with producing a native 4K image, the engine can render the game at a lower resolution and have DLSS make up the difference.

The implications of this technology, especially on computationally limited devices, is immense. For the Switch, which doubles as a battery powered handheld when removed from its dock, the use of DLSS could allow it to produce visuals similar to the far larger and more expensive Xbox and PlayStation systems it’s in competition with. If Nintendo and NVIDIA can prove DLSS to be viable on something as small as the Switch, we’ll likely see the technology come to future smartphones and tablets to make up for their relatively limited GPUs.

But why stop there? If artificial intelligence systems like DLSS can scale up a video game, it stands to reason the same techniques could be applied to other forms of content. Rather than saturating your Internet connection with a 16K video stream, will TVs of the future simply make the best of what they have using a machine learning algorithm trained on popular shows and movies?

Continue reading “AI Upscaling And The Future Of Content Delivery” →

This Negative Reinforcement Keyboard May Shock You

January 28, 2021 by Kristina Panos 23 Comments

We wouldn’t be where we are today without Mrs. Coldiron’s middle school typing class. Even though she may have wanted to, she never did use negative reinforcement to improve our typing speed or technique. We unruly teenagers might have learned to type a lot faster if those IBM Selectrics had been wired up for discipline like [3DPrintedLife]’s terrifying, tingle-inducing typist trainer keyboard (YouTube, embedded below).

This keyboard uses capsense modules and a neural network to detect whether the user is touch-typing or just hunting and pecking. If you’re doing it wrong, you’ll get a shock from the guts of a prank shock pen every time you peck the T or Y keys. Oh, and just for fun, there’s a 20 V LED bar across the top that is supposed to deter you from looking down at your hands with randomized and blindingly bright strobing light.

Twenty-four of the keys are connected in groups of three by finger usage — for example Q, A, and Z are wired to the same capsense module. These are all wired up to a Raspberry Pi Zero along with the light bar. [3DPrintedLife] was getting a lot of cross-talk between capsense modules, so they solved the problem in software by training a TensorFlow model with a ton of both proper and improper typing data.

We love the little meter on the touchscreen that shows at a glance how you’re doing in the touch typing department. As the meter inches leftward, you know you’re in for a shock. [3DPrintedLife] even built in some games that use pain to promote faster and more accurate typing. Check out the build video after the break, but don’t say we didn’t warn you about the strobing lights.

The secret to the shock pen is a tiny flyback transformer like the kind used in CRT televisions. Find a full-sized flyback transformer and you can build yourself a handheld high-voltage power supply.

Continue reading “This Negative Reinforcement Keyboard May Shock You” →

Remoticon Video: How To Use Machine Learning With Microcontrollers

December 4, 2020 by Mike Szczys 5 Comments

Going from a microcontroller blinking an LED, to one that blinks the LED using voice commands based on a data set that you trained on a neural net work is a “now draw the rest of the owl” problem. Lucky for us, Shawn Hymel walks us through the entire process during his Tiny ML workshop from the 2020 Hackaday Remoticon. The video has just now been published and can be viewed below.

This is truly an end-to-end Hello World for getting machine learning up and running on a microcontroller. Shawn covers the process of collecting and preparing the audio samples, training the data set, and getting it all onto the microcontroller. At the end of two hours, he’s able to show the STM32 recognizing and responding to two different spoken words. Along the way he pauses to discuss the context of what’s happening in every step, which will help you go back and expand in those areas later to suit your own project needs.

Continue reading “Remoticon Video: How To Use Machine Learning With Microcontrollers” →

“Enhance” Is Now A Thing, But Don’t Believe What You See

November 25, 2020 by Lewin Day 64 Comments

It was a trope all too familiar in the 1990s — law enforcement in movies and TV taking a pixellated, blurry image, and hitting the magic “enhance” button to reveal suspects to be brought to justice. Creating data where there simply was none before was a great way to ruin immersion for anyone with a modicum of technical expertise, and spoiled many movies and TV shows.

Of course, technology marches on and what was once an utter impossibility often becomes trivial in due time. These days, it’s expected that a sub-$100 computer can easily differentiate between a banana, a dog, and a human, something that was unfathomable at the dawn of the microcomputer era. This capability is rooted in the technology of neural networks, which can be trained to do all manner of tasks formerly considered difficult for computers.

With neural networks and plenty of processing power at hand, there have been a flood of projects aiming to “enhance” everything from low-resolution human faces to old film footage, increasing resolution and filling in for the data that simply isn’t there. But what’s really going on behind the scenes, and is this technology really capable of accurately enhancing anything?

Continue reading ““Enhance” Is Now A Thing, But Don’t Believe What You See” →

Training A Neural Network To Play A Driving Game

November 7, 2020 by Lewin Day 12 Comments

Often, when we think of getting a computer to complete a task, we contemplate creating complex algorithms that take in the relevant inputs and produce the desired behaviour. For some tasks, like navigating a car down a road, the sheer multitude of input data and its relationship to the desired output is so complex that it becomes near-impossible to code a solution. In these cases, it can make more sense to create a neural network and train the computer to do the job, as one would a human. On a more basic level, [Gigante] did just that, teaching a neural network to play a basic driving game with a genetic algorithm.

The game consists of a basic top-down 2D driving game. The AI is given the distance to the edge of the track along five lines at different angles projected from the front of the vehicle. The AI also knows its speed and direction. Given these 7 numbers, it calculates the outputs for steering, braking and acceleration to drive the car.

To train the AI, [Gigante] started with 650 AIs, and picked the best performer, which just barely managed to navigate the first two corners. Marking this AI as the parent of the next generation, the AIs were iterated with random mutations. Each generation showed some improvement, with [Gigante] picking the best performers each time to parent the next generation. Within just four iterations, some of the cars are able to complete a full lap. With enough training, the cars are able to complete the course at great speed without hitting the walls at all.

It’s a great example of machine learning and the use of genetic algorithms to improve fitness over time. [Gigante] points out that there’s no need for a human in the loop either, if the software is coded to self-measure the fitness of each generation. We’ve seen similar techniques used to play Mario, too. Video after the break.

Continue reading “Training A Neural Network To Play A Driving Game” →