Discover more from PATENT DROP
#045 PATENT DROP - Adobe, Nvidia & EA
personalized fashion recommendations, machine vision in games & smart cars
Sorry it’s been a while. I’m currently in LA working at the intersection of music and web3. It’s an early space, but there are interesting opportunities for creating a new decentralised music economy that rewards creators and their fans alike. If you’re interested in this space or in LA, let’s meet!
Before we get into this week’s filings, there’s also a quick message from this week’s sponsors:
How Much $10,000 Invested in TSLA 10 Years Ago Is Worth Now
If you’re anything like me, going back in time and betting the house on TSLA is high on your shoulda-woulda-coulda list. As I’m sure you all heard, Musk’s EV operation recently reached a $1T market cap and joined the ranks of Apple, Microsoft and Amazon.
So what’s a $10,000 investment in TSLA from 2011 worth today? A staggering $1.8M….
But for those of us without a Delorean, alternative assets, like art, can be a great way to seize profits in frothy markets.
We’re talking about a gigantic $6 trillion asset class, loved by tech billionaires like Mr Musk.
Check this out:
Contemporary art prices beat the S&P 500 by 174% (1995–2020)
Total wealth in art is projected to explode by another $1T by 2026
Transaction volume has increased 2,700% since 2010
No wonder the Wall Street Journal declared that “art is among the hottest markets on Earth.”
Thanks to Masterworks.io, NYC’s newest $1B unicorn, you can access the asset class that’s more valuable than Tesla. And, early adopters already got a 32% annualized return from a Banksy painting in 2020.
Why not join them? Patent Drop Readers get priority access*
*See important disclosures
Okay, let’s get into it…
Fashion recommendation systems suffer from a few core problems.
Firstly, with rapidly changing fashion trends, recommendation systems don’t usually have enough data to process and recommend items from new fashion trends.
Secondly, a lot of existing recommendation systems rely on semantic descriptions rather than visual signals to determine what’s fashionable. But there are obviously huge limitations to relying on words to ascertain fashionability.
In this patent application, Adobe reveals that it’s looking at using deep-learning methods that analyse the visuals of fashion items in order to make personalized recommendations to users.
The model will work by studying user behaviour (e.g. which items are they clicking on, or spending more time looking at) and then extracting both the 'latent’ features of the items that are capturing a user’s attention, and the latent features of a user themselves. Recommendations will then be made in a way that matches items that a specific user will find most interesting.
From studying Adobe patents over the last year, it looks like they are doing a lot of work around AI - specifically using deep learning to understand audio, video or photos. On the one hand, these could be useful for making the existing creator tools better. But with something like this fashion recommendation algorithm, perhaps Adobe is looking at leveraging its deep understanding of images into AI models that can be licensed by other companies - e.g. an e-commerce company selling clothes.
If you read #044 PATENT DROP, you would’ve seen how Nvidia is looking at using gaze detection to determine where a driver is looking and then correspond this to any voice instructions given. For instance, if a driver looks towards the sound system and says “turn it up”, Nvidia’s system will detect that the driver is referring to the volume of the entertainment system.
In this filing, Nvidia go a step further and start exploring using speech and image recognition to enable occupants to deliver voice commands in a car.
For instance, if the driver of a vehicle said “lower Sally’s window”, Nvidia’s system will detect where Sally is sat using cameras + image detection, and then lower the window where she’s sat.
Over time, Nvidia would want to build a system that’s capturing vehicle data and learning about its occupants and their preferences over time. This includes creating audio fingerprints so that the speaker is recognised purely on their voice. For example, some commands should only be followed upon if the driver is saying it.
What’s interesting about this play is that by incorporating sensors and microphone data into a car, there are a lot of data points that can be captured. For example, who is a person’s household, what is the close social network around a household (e.g. the friends who appear in the car from time to time), or what are the things they talk about.
EA (developers of FIFA) is looking at using machine vision to improve the reliability of their games.
It’s going to do this by comparing virtual-world sporting events from a game, to data from real world sporting events.
For example, a machine vision model might be able to detect images from gameplay and see if it appears to violate natural laws of physics from real world footage. This could help game developers quickly identify and debug issues in gameplay.
Moreover, with nearly infinite footage of real world events existing online, machine vision could help us develop virtual games that display realistic human behaviour much more easily. For instance, in the filing, EA mention that using real world footage of a player celebrating winning the Stanley Cup could be used to automatically create a similar scene in the virtual world.
As the metaverse begins to take shape and we build out increasingly complex virtual worlds, machine vision will become an essential tool to mimicking the real world in game environments quickly and cheaply. It’s fascinating seeing EA apply this in the context of sports games.
Before you go…
Join 200,000 investors following Patent Drop on public.com for tl;drs of patent summaries on public companies
Know anyone curious about the future of tech? Forward them this email :)