image processing

A six-minute clip of Adobe's MAX 2011 sneak peek showing a Photoshop prototype that removes motion blur from photos. The audience reacts with astonishment as blurry images become legible.

A short explainer from Nat and Friends on where Google Earth's 3D imagery comes from, how it is stitched together, and how often it is updated. It describes aircraft carrying camera rigs and photogrammetry that turns the photos into 3D models.

A short conference talk from !!Con 2017 in which Kevin Chen explains the HDR imaging algorithm from a foundational paper, framing it as a system of linear equations, and implements it in a Microsoft Excel spreadsheet. No prior image processing knowledge is needed.

The first lecture of Stanford's Natural Language Processing with Deep Learning course, taught by Chris Manning and Richard Socher. It introduces NLP and word vectors, including SVD, skip-gram, CBOW, negative sampling and word2vec.

A short video accompanying the CVPR 2018 paper Learning to See in the Dark. It shows a neural network producing sharp images from very low-light raw sensor data, compared against denoising approaches.

A one-minute scene from the film Enemy of the State showing absurdly implausible image enhancement, with freezing, rotating, zooming and focusing on a video.

A SIGGRAPH Asia 2013 research demo of 3-Sweep, a tool for extracting editable 3D objects from a single photo using a few user strokes. It predates modern AI image tools.

A two-minute supercut of film and TV characters ordering a computer to 'enhance' a blurry image into a perfectly sharp one. It is a remaster of an earlier 2009 version.

Andrew Adams, a Halide developer, presents the Halide language for image processing and tensor computation. The talk explains the separation of algorithm from schedule with illustrated examples of parallelism, locality and redundancy tradeoffs.

The Thought Emporium shows how to receive clear images directly from GOES-15, 16 and 17 and Himawari 8 geostationary weather satellites with software-defined radio. It covers the extra effort needed to receive GOES data.

A short Harris Corporation (L3Harris) promotional video showing wide-area airborne motion imagery, a sensor system for persistent real-time aerial surveillance of large areas.

The first lecture of Stanford's CS231n course on convolutional neural networks for visual recognition. It surveys the history and key challenges of computer vision, from the Cambrian explosion and Hubel and Wiesel to ImageNet.

A 2006 Google TechTalk in which Luis von Ahn introduces human computation, using people's effort to solve problems like image recognition that computers cannot yet handle. It covers games with a purpose and related ideas such as CAPTCHAs.

An HBO clip from Silicon Valley season 4 in which the 'Not Hotdog' app is demoed. The app classifies food photos as hot dog or not hot dog.

Adobe's sneak-peek demo of Content-Aware Fill ahead of Photoshop CS5, showing objects being removed and areas repaired seamlessly with the delete key.

The Blade Runner scene in which Deckard uses a voice-controlled machine to enhance a photograph, zooming into hidden details.

MinutePhysics explains that computers store brightness nonlinearly, so blurring and blending colors in raw RGB gives wrong, dark results. It shows why mixed colors, such as in gradients, look off.

Samuel Falvo II live-codes a text preprocessor and blog engine in Forth over about an hour. It shows Forth's dictionary being snapshotted and restored in place of mutation, and how a small Forth system gets real work done.

Mike Pound explains the discrete cosine transform and how it underpins JPEG compression. It is part two of Computerphile's JPEG series.

MIT researchers show that sound can be recovered from silent video by analyzing tiny vibrations in objects such as a potato chip bag. It works even with consumer cameras by using rolling shutter.

A short demo video for NVIDIA's StyleGAN paper, which proposes a style-based generator for GANs. It shows face synthesis with style mixing and scale-specific control over attributes like pose and freckles.

The first lecture in an 18-part series on digital photography from Marc Levoy's Stanford CS 178 course, recorded at Google in 2016. The series covers how digital cameras work and computational photography, with slides and assignments available on the course site.

Jessie Frazelle's DockerCon SF 2015 talk on running Docker containers for desktop applications, with fun container hacks and images. It shows how far containers can be pushed beyond servers.

A King of the Hill clip in which Hank-style character rants about saving images as PNG instead of JPEG after exporting from Windows Paint.

A University of Tokyo lab demonstrates a book scanner that flips pages automatically and scans over 250 pages per minute. It uses real-time 3D recognition of the flipped pages and then restores each one to a flat image.

Berkeley neuroscientist Jack Gallant shows how fMRI brain mapping can be used to decode what a person sees, with visualizations of brain activity and its implications.

A full upload of Alan Kay's 1987 talk, with demos of several Smalltalk-era software systems and ideas about visual, kinesthetic and symbolic reasoning. It includes a Smalltalk program written by a 12-year-old and footage of the Rand tablet.

Technology Connections explains why the superheterodyne radio has its name and how mixing frequencies lets it tune stations well. It walks through the underlying wave phenomena and algebra and why the design solved a hard tuning problem.

David Sachs of InvenSense explains how gyroscopes, accelerometers and compasses work on their own and together, demonstrated on a modified Android handset. He covers sensor fusion algorithms and why rotation math such as Euler angles matters.

An artist describes discovering she has aphantasia, the inability to form mental images, and how it affects her work. She asks viewers about their own visualization.

Neuroscientist Anil Seth argues in this TED talk that perception is a controlled hallucination: the brain constantly predicts the world and our shared agreement is what we call reality. He also touches on the construction of the self.

An NBC Today Show correspondent appears to row a canoe in deep storm floodwater, though the water is only about six inches deep. The uploader notes it is a misleading depiction rather than intentional fake news.

Andrew Ng presents unsupervised feature learning and deep learning, showing how algorithms can learn representations from unlabeled data instead of hand-designed features. He gives examples in vision and audio, including visualizations of features learned by successive layers.

Emily Shea explains how repetitive strain injury pushed her to program by voice using speech recognition (Talon), and gives a live coding demo. She argues voice-driven development can be pain-free, efficient and enjoyable.

Martin Gorner's Devoxx talk introduces deep learning with TensorFlow, walking through neural networks for MNIST, how to choose an architecture and how to diagnose networks that fail to train.

Luis von Ahn, creator of CAPTCHA and reCAPTCHA, explains how small contributions from many people online can digitize books. He then introduces Duolingo as a way to translate the web while people learn languages.

Leo Dirac explains how transformer models have replaced LSTMs in NLP. He gives background from bag of words through RNNs and LSTMs, then a technical deep dive into multi-headed self-attention and positional encoding, with sample code.

Mark Rober and Jabril build an app that uses machine learning to decode baseball sign sequences. The video also explains neural networks in a simple, accessible way.

Peter Norvig's UBC distinguished lecture on how massive amounts of simple data, such as for language modeling, can yield understanding that hand-built models could not.

Rob Miles explains on Computerphile how language models generate plausible text and how transformers improved on recurrent networks, with a look at what comes next.