Laion Big Video Dataset

Laion Big Video Dataset(projects.laion.ai)

90 points by ks2048 3 days ago | 24 comments

"Overall, we attempted to download 130M videos and achieved a link success rate of approximately 60%, resulting in 80M successfully retrieved videos with a total duration of 10M hours."

I am astonished that the success rate is so high. How Youtube didn't block them, I don't know. But I think that this URL list won't age well because youtube will very quickly block any researcher trying to download these videos themselves.

ks2048 3 days ago | |

I believe they have the actual videos to download, "exclusively for academic and non-commercial research." and you have to request access, https://projects.laion.ai/bvd/bvd-raw-access.html

They say they used yt-dlp and "employ a residential proxy network". But, yeah, youtube seems to now more aggressively block yt-dlp.

kennyadam 3 days ago | | |

Ha, is it just me or does a “residential proxy network” sound like a fancy way of saying “botnet”?

kevmo314 3 days ago | | |

That’s a pretty penny on the bandwidth bill alone!

maybewhenthesun 3 days ago | | |

> youtube seems to now more aggressively block yt-dlp.

Another thing to thank AI for! /s

ks2048 3 days ago | |

I was wondering if youtube blocked them a lot and they got a lot of videos from other sources - but no, you can see on this HF page that 93.1% of URls are youtube.com, https://huggingface.co/datasets/laion/BVD-URLs

vivzkestrel 3 days ago |

- can someone with expertise give us an overview of the architecture involved doing this

- let us say you ran yt-dlp inside python aiohttp

- surely your ll run a limit soon as your ip address will be flagged

- what solutions do we have to auto rotate proxies in python

- are there better, faster and more reliable ways to go about downloading a 100 million videos without getting your ip address blocked?

earthnail 2 days ago | |

Use Tor to proxy. Every time you get a connection error, send a SIGHUP to the Tor daemon, which will tell the Tor daemon to establish a new connection, thereby giving you a new IP. You may not get as good results as a residential proxy, but these proxies are prohibitively expensive if you're doing this for your own personal or academic use, and the Tor method works surprisingly well.

vivzkestrel 2 days ago | | |

- am not familiar with tor except for the fact that it uses something called onion with multiple hops and is very hard to trace making it privacy friendly

- are we talking here about using tor programmatically?

- what I am referring to on my end is how to go about programmatically downloading a 100 million videos one after the other on s3

- what kind of bot detection / evasion / ip address / proxy rotation mechanisms are needed

- are you familiar with libraries in the pypi ecosystem capable of tackling some of these concerns?

- is aiohttp / asyncio python enough for this paired with multiprocessing or do we have better options?

m00dy 2 days ago | | |

Dude, Tor's exit nodes are flagged everywhere.

faangguyindia 3 days ago | |

Google may supply cache boxes to ISPs which can be manipulated to let download any video you want

voidUpdate 3 days ago |

Damn, that's a lot of videos for them to contact the creators and ask for permission to use their content as machine learning training content. Unless of course, they didn't, and just went ahead with it anyway...

lucrbvi 3 days ago | |

Under the EU's AI Act non-profits and universities can just bypass these constents to make non-commercial datasets btw

Someone 2 days ago | | |

I don’t see the AI act giving those permissions. In fact, I don’t think the AI act applies at all here.

https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:...:

“This Regulation does not apply to AI systems or AI models, including their output, specifically developed and put into service for the sole purpose of scientific research and development.”

https://github.com/LAION-AI/BVD:

“LAION-BVD is released exclusively for research purposes and not for commercial use”

voidUpdate 3 days ago | | |

Does that apply to everyone who doesn't live in the EU as well? And what happens when a major AI company uses this dataset as training data, ignoring consent as usual?

stymaar 3 days ago | | |

But I've been told that the AI act was a terrible anti-innovation legislation made by clueless and corrupt bureaucrats. How is it possible that they made a sensible decision?

techsage 1 day ago |

Curious how useful very large video datasets are compared with smaller, better curated ones.

At some point, adding more videos probably helps less than improving the quality of the data.