Clever Camera App Uses Deep Learning to Perfectly Retouch Your Photos Before You Take Them

  • 03/08/2017

Professional photographers often spend hours painstakingly perfecting their images in Lightroom and Photoshop before sharing them with the world. But researchers at MIT are promising similar results generated so quickly that your smartphone can correct and retouch a photo before you’ve even taken it.

As with many image-processing breakthroughs we’ve seen over the past few years, the ‘secret sauce’ of this new app lies with an artificial neural network that’s able to learn how to retouch photographs using before and after examples. In this case, researchers used a data set they created in cooperation with Adobe Systems which included 5,000 images that had each been color-corrected and retouched by five different professional photographers, allowing the neural network to generate its own guidelines on what the final result should look like given a specific starting image.

But training the software how to automatically correct and improve photos was only half of the technical achievement here. We’ve seen examples of modifying photos so they match the style of another photographer before, but they typically rely on the cloud, sending images to a more powerful computer that handles the actual processing. That takes time, however, and would require a smartphone to have a reliable data connection in order for it to work—and that’s not always the case.

The researchers at MIT, working with Google, came up with a way for all of the processing to happen within the standalone app, even on a smartphone with dated specs. Previous approaches performed the retouching on a low-resolution version of the source image to speed up the process, but trying to up-res the results inevitably produces terrible looking photos. So instead of generating a tiny sample using the machine-learning system, the app creates a set of formulae that can be used to quickly process all of the pixels in a high-def version of the source image.

How quickly? The new approach allows for these color corrections to be made in real-time. So as you’re framing a photo using your smartphone’s screen, before you even hit the shutter button, you can see what the results would look like when processed to match another photographer’s style. Then once you snap your photograph, the full-res version can be quickly processed by the app, though not in real-time given all the extra pixels.

There’s no word on when the researchers plan to release their app to the public, or if the technology will be incorporated into existing apps from Google. The best possible outcome would see this technology licensed out to other camera app makers, allowing complex filters to be previewed in real time. Imagine having access to all of Instagram’s filters while you’re photographing your next fancy dessert, letting you perfect the framing and colors so you can ensure you’ll get as many ego-stroking ‘likes’ as possible.

Source: Gizmodo

Share to:

Comments

Top Mobile App Development Trends in 2021

It’s no secret that mobile apps are among the fastest-growing sectors of the world’s technology infrastructure with 2.1 million Android apps and 2 million iOS apps now available on leading app stores. With a diverse array of use cases encompassing everything from direct-to-consumer service delivery to business-to-business communication.

  • 15/02/2021

10 Apps To Connect Your PC to Android Devices

When you’re a vivid smartphone user, you may need to connect your device to your PC or Mac. Maybe you wish to backup your photos and videos, extract some crucial data, or extract files to free space on your device. Or maybe you wish to check your calls and messages or remote-control your device.

  • 01/02/2021