Comment by gps372 | Hacker News Reader

Comment by gps372 | Hacker News Reader

Parent Full thread

Most comments may have ignored this so far, but this is an interesting AI based research.

At the end of the article is this excerpt

>> Results from this research were generated using an AI computer vision model developed by the research team on Bristol’s Isambard-AI, the UK’s most powerful AI supercomputer. Isambard-AI powered the analysis of 172.6 hours of live match footage from all 104 games played between 11 June and 19 July 2026.

So, basically now merchandizers, auditors, compliance monitors, etc. have a tool using which they can quantify whether broadcaster complied with contractual requirements of showing their ads/merch/logos as per contract. And if they missed, then they can quantify the delta - to sue the broadcaster for the balance as well.

Also, it can enable many countries, where certain content is not allowed, then this tool can simply give them data points without having to employ someone, or have someone review the output of the AI tooling.

I'm sure this was already easily doable for years by using plain and boring computer vision

It was easily doable for even more years using plain and boring MK1 eyeballs and a stopwatch.

If you are spending millions on ads, you can pay a couple of guys to watch TV and take notes.

Broadcasters also keep logs of their broadcasts for verification purposes. The traffic systems that inject the ads have parameters for how often and what times certain ads are played which are quite deterministic.

For the World Cup, this was always easy.

It was not easy for your local triple A baseball game being streamed to 500 viewers. This approach scales down.

Hours of bleeding-edge CV analysis sounds pretty expensive for a triple-A game streamed to 500 pairs of eyeballs. They’ll be using the pen & paper approach, too.

Not at scale - a thousand games automatically scored: easy, just feed in the new video files. A thousand workers for individual matches across the country and different channels: absolutely not.

Not to mention human accuracy will likely be way, way worse, no human being will maintain their attention counting ads at five second intervals for even ten minutes, almost two hours is unthinkable.

If attention is all you have, humans are pretty bad at it for long boring tasks.

Since people have bought air time on a broadcast, people have done this very thing to ensure their ad ran. They also notice other analytics like what other ads ran around it and what the content of the program was when it ran. Can't run an ad selling a new car when the scene just before the break was a brutal car wreck or something.

Eventually, people buying ad time on TV stations could request an air check and receive a VHS tape showing a bit of the programming before/after the ad pod your ad ran. Particularly useful when buying ad time on local stations out of the market where you lived.

I remember google bought a company that recorded local FM broadcasts around the country to independently verify ad plays.

I worked for a company in 2020 that did ad analysis with computer vision - specifically in sports, quantifying value for sponsors based on exposure on team socials (accounting for logo size, positioning, etc in photos + videos) - I was a humble annotator https://www.blinkfire.com

Plain and boring computer vision would have still required a team who would identify frames to sample before processing, and then annotate, generalize and classify the data.

Not an expert on this but I don't think same kind of manual effort is required anymore.

It was never required because you already had the frames labeled. Because you gave the ad footage to the station. You created the ad, you already have a golden copy, you just need to verify N sequential frames match to within X percent of pixels.

> by using plain and boring computer vision

Technology is funny, I remember when an intern project detected cats in Youtube videos and it was front page news (round here, at least).

Now it's just boring technology.

>"You're flying! You're sitting in a chair, in the sky!" --PP toucher CK.

This goes along well with another saying I have.

"You will never prove machines are intelligent, you will only prove humans are not"

You probably don't even need vision. The audio on adverts is noticeably different.

There's no audio on the adverts in question.

> The analysis counted advertising built into the match footage (e.g. on the hoardings), rather than commercials shown during advertising breaks

This is also being done in retail, where manufacturers and/or distributors pay for preferred placement (end caps, eye level, etc), but how do you check that e.g. Safeway is really doing that across a couple of thousand stores?

You hire minimum wage people to walk around stores with cameras. This was actually done with notepads occasionally, but so much less expensive tondo with AI.

True, in-store planogram compliance and floor-plan compliance have got in focus again. It never lost the focus to be honest, just that now there is an expectation to do it while ensuring nothing fall through the cracks.

But there is a much more stringent privacy requirement here since customers walking in those stores haven't really signed up for this analysis. Hence, difficult to just setup the camera and take continuous data for analysis.

Interesting on the privacy angle. I’m not sure anyone is really caring about that; there are lots of companies offering automated compliance services. I’m not super close to that industry but close enough that I’ve heard the pitches, and even the “no expectation of privacy in public” CYA has never come up.

What’s going to be fun are robotic stockers, racing to ensure compliance when a person/bot flagged as an auditor is in the store.

You’ll be reaching for the Pepsi and some robot will swoop in and replace the entire shelf with Coke when the Pepsi auditor walks into the store. Then you’ll reach to bottom shelf for your Coke and just as the robot swaps it back to eye level because the Coke auditor walked in. The store having sold the same placement to both.

(Ok, a bit far-fetched, but entertaining)

This might be already happening. Competitor's market share (without details) is common to share. Scintilla onboarded by Walmart earlier this year might already be giving instore AI capabilities.

>since customers walking in those stores haven't really signed up for this analysis

In the US at least there is no requirement for this. You're already being watched, tracked, and put in identification databases as it is.

There's a lot of data extraction being done for every game. It already uses tons of AI, and has been using it for years. Doing it by hand would have been more surprising.

What's also surprising is that despite all this data we can't automate those soccer player jobs yet.

> So, basically now merchandizers, auditors, compliance monitors, etc. have a tool using which they can quantify whether broadcaster complied with contractual requirements of showing their ads/merch/logos as per contract

Automated ad-monitoring tools existed for decades for TV and Radio using even older DSP techniques and imperceptible audio/visual watermarks.

Reply on news.ycombinator.com