image processing

A guest video by Welch Labs on 3Blue1Brown explaining how text-to-image and video models work. It covers CLIP, shared embedding spaces, diffusion models (DDPM, DDIM), conditioning, guidance and negative prompts.

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 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.

GeoGuessr pro Rainbolt plays head-to-head against an AI built by Stanford students that guesses locations from street-view images. The video shows the AI repeatedly beating him.

Alex O'Connor shows that ChatGPT cannot draw a wine glass filled to the brim, and ties the failure to David Hume's missing shade of blue. It asks what the failure says about how image models learn.

Benn Jordan's video about storing a PNG image in a living bird, a starling that learns to reproduce encoded sounds. It mixes ornithology, audio engineering, and data encoding.

Peter Norvig, then Director of Research at Google, talks at Startup School 2008 about leveraging large amounts of data and about technical challenges in Google projects such as image search. A central idea is that data can beat hand-written programming.

Veritasium visits the US Navy's Crane facility to test top night vision goggles. It explains how image intensifier tubes and thermal or digital systems work and compares the generations of the technology.

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.

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.

Jim Keller's talk at the 2020 Scaled Machine Learning conference on Moore's Law in the age of AI chips. He compares scalar, vector and GPU-style parallel approaches to computation.

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.

Andrej Karpathy, then Tesla's head of AI, presents at ScaledML 2020 on how Tesla builds full self-driving using vision-based neural nets and fleet data collection. He covers training custom detectors and neural nets gradually replacing traditional code.

Benn Jordan explains how Flock Safety's license plate cameras log and share vehicle locations, and explores how adversarial AI can defeat them. It covers weak security and the surveillance risks of private police-camera networks.

David MacKay's 2007 Google Tech Talk presents Dasher, a text-entry system driven by continuous gestures and a language model, reaching about 35 words per minute with one finger and 25 hands-free. It explains the information-theory reasoning behind the design.

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.

Andrej Karpathy, then Tesla's head of AI, gives a CVPR 2021 workshop keynote on how Autopilot's neural networks work. A large part covers the move to a vision-only stack, including why radar was dropped and how vision produces depth estimates.

Peter Norvig's 2007 Google talk on theorizing from data, covering what huge datasets can do for language understanding, translation, information extraction and inference. It argues that more data often beats cleverer models.

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.

A PyCon 2012 talk by Kurt Grandis on building an automated water-squirting turret that uses Python and computer vision to detect and drive off squirrels from a bird feeder.

Pokey Rule presents Cursorless, a spoken language designed for editing code by voice, built on tree-sitter parsing with smart structural and simple token-level primitives. The talk is from Strange Loop 2023.

Mobileye co-founder and CTO Amnon Shashua explains what goes into sensing for autonomous driving and whether end-to-end deep learning can work for it. He argues for decomposing the problem.

Andrej Karpathy's CVPR 2021 presentation on Tesla's Autopilot, covering its vision-based approach, sensor choices and the scale of Tesla's deep learning compute cluster. This upload is a lower-quality copy of the talk.

Tom Gally, a former professional translator and University of Tokyo professor, discusses in April 2023 how GPT-4 and similar models may affect translators' careers. He draws on Japanese-English translation experience.

A HOPE XI talk on programming by voice with open source speech recognition. The speaker writes Linux systems code by voice and describes building custom voice grammars as an alternative to the keyboard for programmers with repetitive strain injuries.

Martin Kleppmann argues that databases are global shared mutable state and proposes treating them as an ever-growing log of immutable facts. Derived data such as caches, indexes and views is then built by stream processors like Apache Samza. The talk draws on Kafka and LinkedIn's experience with commit logs.

Reducible's visual walkthrough of the Fast Fourier Transform, motivated by polynomial multiplication. It shows how the FFT's key ideas can be discovered by asking the right questions.

Veritasium tells the story of the Fast Fourier Transform, from its use in detecting nuclear tests to Cold War politics, and how it could have changed the arms race. It includes a short, intuitive explanation of what the Fourier transform does.

Robert Erdmann, a materials scientist at the Rijksmuseum, gives a PyCon keynote on the ultra-high-resolution imaging of Rembrandt's Night Watch. He covers the automated gantry camera system, the Python software that drives it, and the large volumes of image data involved.

Photopea founder Ivan Kutskir talks about developing a free browser-based photo editor, one of the most complex JavaScript programs ever written, and the web technologies behind it.

David Kriesel's FrOSCon talk on how Xerox scancopiers, due to JBIG2 compression, silently replaced digits in scanned documents. He walks through how he discovered the bug and why it is so dangerous.

Michael Levin's NeurIPS 2018 talk on bioelectric computation outside the nervous system, showing how cell electrical networks shape development and regeneration. It covers morphological information processing in living tissue.

Inigo Quilez live-codes a cartoon character painting entirely from math, using signed distance fields in a Shadertoy shader. He explains each step of building the image from formulas.

Rick Houlihan's AWS re:Invent 2018 deep dive on advanced DynamoDB design patterns. It covers index overloading, GSI sharding, composite keys, graph processing and transactions.

Explains the proposed NASA JPL plan to use the Sun as a gravitational lens, sending solar sail craft to its focal region to image an exoplanet's surface. It covers the distances involved, how imaging with the Sun would work and the mission design.

Ben Eater explains how VGA signals work and builds a breadboard circuit that produces correct sync timing so a monitor recognizes the signal. He then builds toward displaying an image from simple logic chips.

PBS Space Time explains the proposal to use the Sun's gravity as a giant lens, sending a train of small light-sail spacecraft to the focal point roughly 550 AU out. It covers how such a telescope could image exoplanets in surprising detail and the engineering challenges involved.

Myst co-creator Rand Miller recounts how the HyperCard-built game struggled to run on 1x CD-ROM drives. He and his brother Robyn compressed image and audio data heavily to get acceptable performance. Part of Ars Technica's War Stories series.