🍊 The Juice: Dynamic Data

Zumo Labs presents The Juice, a weekly newsletter focused on computer vision problems (and sometimes just regular problems). Get it while it’s fresh.

Week of March 29–April 2, 2021

____

It may be April, but this is no joke: zpy, the open source synthetic data toolkit from Zumo Labs, is now available on GitHub. Click through and you’ll find everything you need to get started making your own synthetic training data for computer vision.

Tools like zpy will be a game changer for folks that have thus far been stuck training their models on static datasets. The iterative nature of synthetic data makes for dynamic datasets, giving the end user fuller control over their entire machine learning stack. We get excited just thinking about it, and we actually just published a piece collecting our thoughts on how open core products like ours fit into the future of computer vision.

Now, onto the roundup. As always, if you enjoy this week’s newsletter, please forward it along to the nerds in your life.

____

#MTurk

Is there any product more appropriately named than Amazon’s Mechanical Turk? Named for the late 18th-century chess-playing automaton that was in actuality operated by a hidden human, the platform allows folks to outsource work (often jobs considered menial or tedious) to a network of workers. This approach is standard operating procedure for content moderation and most data labeling operations. The problem with it, this piece posits, is that “unseen labor” often results in forgotten laborers.

AI: Ghost workers demand to be seen and heard, via BBC.

#Eleuther

The code for GPT-3 has not been released, in part because OpenAI is betting that they can commercialize it. But that’s not stopping a collective called Eleuther from attempting to create an open source version of the language model. It seems like mitigating some of the known issues and biases in GPT-3 is one goal of their work: “The data set that Eleuther is using is more diverse than GPT-3, and it avoids some sources such as Reddit that are more likely to include dubious material.”

This AI Can Generate Convincing Text — and Anyone Can Use It, via Wired.

#RoboCoffee

If you make pour-over or espresso at home, you understand that a degree of mechanical precision is required to nail the perfect cup. That may be why so many outfits are aiming to bring robot baristas to market. Jarvis, pictured below, is a robo-barista from Blue Hill Coffee. What sets Jarvis apart is the computer vision tech he’s fitted with, which allows him to both train faster and be adaptable when things aren’t where they should be. Can a robot replace the warmth of a brief human interaction at the coffee shop? Maybe… if it can spell my name right.

Blue Hill Brings Computer Vision to Its Coffee Robot to Recreate the Barista Experience, via The Spoon.

#Infrastructure

While city planners do their best to consider how the public will use the spaces they create, they don’t always get it exactly right. In some cases, that’s just because things change — such as the unprecedented spike in e-commerce and accompanying fulfilment logistics over the past couple of years — but it can also be due to a lack of accurate data. In this piece from Harris Lumis, the CEO of Automotus, he outlines some of the problems that they’ve been able to solve with computer vision technology.

Computer vision software has the potential to reinvent the way cities move, via TechCrunch.

#TrainingData

A company called Cognitive Pilot is outfitting Russian trains with computer vision technology. The systems are intended to detect obstacles on the track and alert the crew in an effort to reduce railway accidents. While there’s no timeline provided for this pilot program, the trains may one day be operated entirely by AI. That gives new meaning to the term computer vision engineer.

Russian trains to be equipped with computer vision, via Reuters.

#ThisOldHouse

If you’re a homeowner, you probably wrestle with the question of which home improvement projects are actually worth it. A new app that launched for iOS this month, Plunk, claims to be able to answer that question using computer vision. The press around the release is light on details of their tech, but the app apparently performs image analysis for real time home valuation and then recommends specific remodeling projects that will have the biggest impact on the home’s value.

This New App Is Designed to Predict Which Reno Projects Will Improve Your Home’s Value, via BobVila.com.

____

📄 Paper of the Week

Factors of Influence for Transfer Learning across Diverse Appearance Domains and Task Types

Transfer learning is standard practice in the computer vision world, but unfortunately it is still very much in the realm of alchemy. This paper performs over 1200 transfer experiments with a large variety of image domains and tasks. It confirms (with science!) some of the best practices we have all learned over time: make sure the image domain for the source dataset is similar and/or broader than that of the target dataset. The growing size of datasets and the increasing cost of training on those massive datasets means transfer learning is here to stay, so understanding the way it works is indispensable for any ML practitioner.

____

Think The Juice was worth the squeeze? Sign up here to receive it weekly.