5 ms·

Unlogo project: remove logos from videos

27 points by xtacy 16y ago

zalew 16y ago
A few months ago we ran into such concept while chatting with a fellow programmer. But instead, we thought about using such filter to count appearances of brands in youtube videos. If their product works I bet this dudes will be bought by some corporateion and their app/filters will be used for sth like that.
- pbhjpbhj 16y ago
  >If their product [unlogo] *works I bet this dudes will be bought by some corporateion and their app/filters will be used for sth like that. Erm, I bet it will be used to replace company logos/brands with ads/logos of other companies that have paid the video distributor (YouTube FTW!) instead. Indeed I'd be highly surprised if this were not the goal all along.
  - zalew 16y ago
    Indeed I'd be highly surprised if this were not the goal all along. Yeah, I didn't want to put it that way...
    - jefftimesten 16y ago
      Jeff Crouse (creator of Unlogo) here. I can guarantee you that is not the goal. I have been working on open source, hacktivist projects for over 5 years. Please see my website: http://www.jeffcrouse.info http://www.jeffcrouse.info
- trin_ 16y ago
  google would be a good fit. analyse the youtube library and sell the stats to the companys. but i guess they're already working on something like that. if not someone in mountainview owes me a beer.
Kliment 16y ago
There was this idea floating around on some virtual reality forums some years back about an "adblock for real life" that would work in a very similar way with a head-mounted display. In terms of pattern recognition it is not at all difficult. Detection and tracking of a set of images is pretty much a solved problem in computer vision (doing it reliably in realtime is still somewhat of a black art). I like how this is framed as an art project, and it's a very cool proof of concept, though it's an issue that it doesn't black out logos until after they have been seen, and the blackout box blinking on and off as the detector passes its threshold only draws attention to those areas. A practical implementation could use inpainting algorithms (also built into OpenCV, and even provided in one of the samples) to fill the logoed areas with background-like texture, so that it's not an obvious "censored" black patch but a smooth transition.
- thesystemis 16y ago
  see also the artvertiser, by julian oliver, damian stewart and more: http://selectparks.net/~julian/theartvertiser/ http://selectparks.net/~julian/theartvertiser/ which is an art project (along similar lines as unlogo) which replaces advertisements with art. nice to see both unlogo and artvetiser be open source, it will be interesting to see the different directions people can take these ideas.
  - jefftimesten 16y ago
    Yes -- Julian's project is great, and was definitely an inspiration to me. It is mentioned in the video: http://www.kickstarter.com/projects/816924031/unlogo-the-corporate-media-filter-0 http://www.kickstarter.com/projects/816924031/unlogo-the-cor...
- chopsueyar 16y ago
  I believe Limor Fried created some type of glasses that would detect refresh rates of video sources and black out the glasses when viewing the monitor/tv. No image recognition, though.
  - TeMPOraL 16y ago
julian37 16y ago
This reminds me of a fairly simple approach to commercial skipping [1] in the early days of DVRs: the TV station's logo is never visible during commercials but always during regular programming. This was for German TV in the 1990s/early 2000s. I've never watched much TV, neither in Germany nor elsewhere, so I don't know if this applies to TV in other countries or if it still applies in Germany these days. Still, I always thought this was a smart solution to an otherwise quite difficult problem. [1] http://en.wikipedia.org/wiki/Commercial_skipping http://en.wikipedia.org/wiki/Commercial_skipping
mtw 16y ago
implementation is ugly
chopsueyar 16y ago
Has it been tested on this? http://en.wikipedia.org/wiki/Logorama http://en.wikipedia.org/wiki/Logorama http://vimeo.com/10149605 http://vimeo.com/10149605
danudey 16y ago
I'd originally thought that the project was to remove those faded 'station id' logos that TV channels put on videos. It would be great if there were some kind of functionality in ffmpeg to remove those semitransparent logos either when the video is first being encoded, or when it's being decoded later to be displayed.

27 points by xtacy 16y ago

zalew 16y ago

A few months ago we ran into such concept while chatting with a fellow programmer. But instead, we thought about using such filter to count appearances of brands in youtube videos. If their product works I bet this dudes will be bought by some corporateion and their app/filters will be used for sth like that.

pbhjpbhj 16y ago
>If their product [unlogo] *works I bet this dudes will be bought by some corporateion and their app/filters will be used for sth like that. Erm, I bet it will be used to replace company logos/brands with ads/logos of other companies that have paid the video distributor (YouTube FTW!) instead. Indeed I'd be highly surprised if this were not the goal all along.
- zalew 16y ago
  Indeed I'd be highly surprised if this were not the goal all along. Yeah, I didn't want to put it that way...
  - jefftimesten 16y ago
    Jeff Crouse (creator of Unlogo) here. I can guarantee you that is not the goal. I have been working on open source, hacktivist projects for over 5 years. Please see my website: http://www.jeffcrouse.info http://www.jeffcrouse.info
trin_ 16y ago
google would be a good fit. analyse the youtube library and sell the stats to the companys. but i guess they're already working on something like that. if not someone in mountainview owes me a beer.

Kliment 16y ago

There was this idea floating around on some virtual reality forums some years back about an "adblock for real life" that would work in a very similar way with a head-mounted display. In terms of pattern recognition it is not at all difficult. Detection and tracking of a set of images is pretty much a solved problem in computer vision (doing it reliably in realtime is still somewhat of a black art). I like how this is framed as an art project, and it's a very cool proof of concept, though it's an issue that it doesn't black out logos until after they have been seen, and the blackout box blinking on and off as the detector passes its threshold only draws attention to those areas. A practical implementation could use inpainting algorithms (also built into OpenCV, and even provided in one of the samples) to fill the logoed areas with background-like texture, so that it's not an obvious "censored" black patch but a smooth transition.

thesystemis 16y ago
see also the artvertiser, by julian oliver, damian stewart and more: http://selectparks.net/~julian/theartvertiser/ http://selectparks.net/~julian/theartvertiser/ which is an art project (along similar lines as unlogo) which replaces advertisements with art. nice to see both unlogo and artvetiser be open source, it will be interesting to see the different directions people can take these ideas.
- jefftimesten 16y ago
  Yes -- Julian's project is great, and was definitely an inspiration to me. It is mentioned in the video: http://www.kickstarter.com/projects/816924031/unlogo-the-corporate-media-filter-0 http://www.kickstarter.com/projects/816924031/unlogo-the-cor...
chopsueyar 16y ago
I believe Limor Fried created some type of glasses that would detect refresh rates of video sources and black out the glasses when viewing the monitor/tv. No image recognition, though.
- TeMPOraL 16y ago

julian37 16y ago

This reminds me of a fairly simple approach to commercial skipping [1] in the early days of DVRs: the TV station's logo is never visible during commercials but always during regular programming. This was for German TV in the 1990s/early 2000s. I've never watched much TV, neither in Germany nor elsewhere, so I don't know if this applies to TV in other countries or if it still applies in Germany these days. Still, I always thought this was a smart solution to an otherwise quite difficult problem. [1] http://en.wikipedia.org/wiki/Commercial_skipping http://en.wikipedia.org/wiki/Commercial_skipping

mtw 16y ago

implementation is ugly

chopsueyar 16y ago

Has it been tested on this? http://en.wikipedia.org/wiki/Logorama http://en.wikipedia.org/wiki/Logorama http://vimeo.com/10149605 http://vimeo.com/10149605

danudey 16y ago

I'd originally thought that the project was to remove those faded 'station id' logos that TV channels put on videos. It would be great if there were some kind of functionality in ffmpeg to remove those semitransparent logos either when the video is first being encoded, or when it's being decoded later to be displayed.