we stopped watching tv in our home a few years ago, coupled with pi hole and ublock origin across all browsers, i personally do not see ads and neither does most of my family. if i do see them at a friends house or somewhere, it is a really jarring experience.
i was at a friends house recently and they were watching tv. i found the ad break as a horrifying experience. the volume is turned too high, you can't skip it and it becomes irritating after the third ad break in an hour.
In 2012, USA tried to implement the CALM act, which mandated content and ads to be at the same level (“A/85”), but it didn’t play out because ads are measured individually whereas a program is measured as a whole. Complaints to the FCC are still about the loudness of ads… See https://www.izotope.com/en/learn/the-mixers-guide-to-loudnes...
A while back I built an ad/not-an-ad classifier for TV frames based purely on pixels (i.e. not prior knowledge of specific ads). One of the interesting findings -- perhaps obvious in retrospect -- was that ad frames were on average brighter than content frames.
I’m sure it’s out there, but I had a similar idea to make an audio device plugin like soundflower to pipe audio through and automatically mute when it detects an ad. We really need an OpenWrt-like project for smart TVs so that you could run something like what you created to mute the sound and display art during ad breaks.
How accurate did you get? Are frame-to-frame probabilities pretty independent? I wonder if you can hit some arbitrary threshold by implementing something like "n frames in a row."
The time series of frame probabilities is of course highly autocorrelated, but I never bothered to model that due to lack of interest. From memory, a simple single-frame classifier was >95% accurate (this was quite a few years ago, mind you). In my small-sample and not very rigorous testing, this was higher than human rates. Turns out that on average it's pretty hard to tell an ad from content just by looking at the pixels of a single frame.
I am quite confident that it wouldn't be too hard to build an ad detection ML model that would have near-perfect accuracy. That said, an approach based on algorithmically detecting repeated segments of lengths consistent with ad spots would work just as well, if not better.
P.S. One thing I thought was really interesting was that the classifier -- that was only ever shown a binary label (ad/not-an-ad) -- learnt an embedding that grouped together entire categories of things across TV networks and geographies (studio news, weather, traffic reports etc).
I don't know much about TV broadcasting, so sorry if this is dumb, but what's a segment length? I assume it is either: lengths of a particular shot, or (probably more likely) it is some extra information that is broadcast with the TV signal about how long a chunk of broadcast is?
I like the idea of looking at pixels, just because that's the sort of info that gets sent down the HDMI cable and will always be available.
There exist inline insertion signalling standards (e.g. SCTE-35). These could be used for the insertion of ads. However, none of this signalling typically makes it into the final over-the-air or cable broadcast and so is not useful for ad detection.
To your question on segment lengths, ad spots have specific, predefined duration. In the US these are typically 15s, 30s and 60s (sometimes 45s). This property could be exploited to detect ads. Consider, for example, a video segment that's exactly 30s in duration and is repeated many times over multiple TV channels. It is very likely to be an ad.
