Watch how machine learning can enhance low-light images

At this year's Conference on Computer Vision and Pattern Recognition, researcher Chen Chen presented a cool project that vastly improves the quality of images captured in low-light conditions.

Via his presentation:

Imaging in low light is challenging due to low photon count and low SNR. Short-exposure images suffer from noise, while long exposure can induce blur and is often impractical. A variety of denoising, deblurring, and enhancement techniques have been proposed, but their effectiveness is limited in extreme conditions, such as video-rate imaging at night. To support the development of learning-based pipelines for low-light image processing, we introduce a dataset of raw short-exposure low-light images, with corresponding long-exposure reference images. Using the presented dataset, we develop a pipeline for processing low-light images, based on end-to-end training of a fully-convolutional network. The network operates directly on raw sensor data and replaces much of the traditional image processing pipeline, which tends to perform poorly on such data. We report promising results on the new dataset, analyze factors that affect performance, and highlight opportunities for future work.

Here's the full project page for more information.

Let's enhance!

CVPR 2018: Learning to See in the Dark (YouTube / Chen Chen) Read the rest

Artist runs classic oil portrait through an algorithm for a cool effect

Artist Dimitris Ladopoulos ran a favorite Rembrandt painting through an algorithm he's used for abstract work and generated a cool RGB-value dimensional output that resembles a blocky paint by numbers. Read the rest

AI is taking photo enhancement to new extremes

Single image super-resolution (SISR) is an emerging technology that uses automated texture synthesis to enhance dithered and blurry photos to nearly pristine resolution. This example from EnhanceNet-PAT shows one type. There's even a free website called Let's Enhance where you can up-res your own images. Read the rest

This algorithmic generative art explores the visual beauty of math

Romanian artist HyperGlu creates programs and algorithms that generate fascinating images and animations with a geometric and mathematical beauty. Read the rest

Photochrome: keyword-generated impressionistic color palettes

Photochrome is a nifty algorithm that generates a color palette based on a keyword you enter. It compiles all images in their database tagged with your keyword and averages the results into RGB and hex values. Read the rest

Solo Radio uses AI to match songs to your facial expression

At this week's London Design Festival, design firm Uniform displayed Solo Radio. Stand in front of the device and it scans your face for input into software that assesses your emotions. Then it plays a song via Spotify algorithms with the appropriate mood. Read the rest

Dance your Ph.D. thesis: Teaching a robot to appreciate beats

Every year, intrepid Ph.D. students face off in a high-stakes competition for honor, glory, and the intermingling of science and art. The goal: Dance your Ph.D. thesis. I showed you the finalists last year. This year, Science magazine has posted all 53 entries online, before the finalists are chosen. I'll confess, I've not yet watched them all. So I can't say this is my favorite, but it is well-done and did immediately catch my attention.

"Human-Based Percussion and Self-Similarity Detection in Electroacoustic Music" is, basically, researcher J. Anderson Mills' attempt to teach a computer to hear percussion sounds the way a human does. In the video, Shiny Robot learns how to dance. You can read a full description of how the various parts of this dance tie into Mills' research at the video site:

The dissertation research began with a two-choice, forced-interval experiment in which 29 humans were asked to rate isolated sounds from most to least percussive. The sound characteristic of rise time was found to be the most correlated with percussion of the characteristics tested. The experiment is represented in the dance by the first two interactions between Alain and Shiny, during which Shiny expresses his inability to correctly choose the stronger percussion sound.

... The final stage of the dissertation research was to use the detection algorithm with real-world music to discover self-similarity in the percussion patterns. By using auto-correlation analysis, the detection algorithm can be used to time the repetition and near repetition in music percussion.

Read the rest