LCLV: Real-time video analysis with Moondream 2B & OLLama (open source, local). Anyone want a set up guide?

[-]

InterstellarReddit@reddit

What would be the best way to do saved videos vs real time using this? I have some old videos that I would love to run though this and see how it behaves.

[-]

Hunting-Succcubus@reddit

very useful to detect slave's i mean employee's emotion or fatigue level.

[-]

And they don’t even need a large model to achieve it. I hope the eventual regulators take note that it’s the applications which are potentially harmful, not the number of gpu it uses, or size, or number of weights.

Once again it’s how evil people can use something that is the problem rather than the thing itself.

[-]

hyperdynesystems@reddit

BRB making this into a commercial software to dunk on Amazon software engineers as hard as possible in the most draconian way so that Amazon gets shut down after no one wants to work there.

Only half kidding, I guarantee they'd buy this given they already use the "snitch on your coworkers" app for their engineering departments lmao.

[-]

SkepticScribe@reddit

Amazon wants a workforce that doesn't need breaks, doesn't get tired, and certainly doesn't bitch about working conditions—including being constantly monitored.

That’s why over the past few years, they’ve been swapping out human workers for advanced AI-driven robots. Currently they “employ” over 750,000 of them! If you think that’s just Amazon's little secret, think again. Other companies are salivating at the cost savings and will most certainly jump on this bandwagon.

[-]

hyperdynesystems@reddit

No wonder their service sucks

[-]

bidet_enthusiast@reddit

Yes please!

[-]

ParsaKhaz@reddit (OP)

https://www.reddit.com/r/Moondream/s/Qn70IPqUez

Would you prefer a video?

[-]

bidet_enthusiast@reddit

No. I prefer written tutorials, but a supplementary video is sometimes nice to have.

[-]

nokia7110@reddit

Yes please to the video and great work btw

[-]

mace_guy@reddit

Isn't the analysis completely wrong. For the same scene, its giving Male, Female and both.

[-]

Correct_Key_7623@reddit

The respon had a slight delay of responding to the ui, you can check at the timeframe.

[-]

hyperdynesystems@reddit

No one's going to comment on its hydration analysis of the baby lol.

> Baby's skin looks dry and flaky

WUT XD

[-]

AnonsAnonAnonagain@reddit

This looks really cool! 🤯

[-]

cddelgado@reddit

Do you realize what you've done? I don't think you do.

The Americans with Disabilities Act requires WCAG 2.1 AA (a web standard) compliance for all publicly available information used by federal, state, and local government agencies, like universities. That WCAG 2.1 AA standard requires separate audio description to be added to videos. A person talks, a scene changes to invoke an emotion or communicate a detail, and there is supposed to be a voice laid on top of the audio track that describes those meaningful changes.

Your utility goes a long way towards creating that. Now, companies offer services for it, but it is highly cost prohibitive. Your tool is *not* cost prohibitive.

To do this well, multiple passes over the video is needed, but all the tools to make automated video description exists. The hardest part will be the last 20% by finding the meaningful expressions, then overlaying the voice in a smart way.

But you took a huge bite out of that apple.

[-]

ParsaKhaz@reddit (OP)

We are actually working on releasing a recipe for video captioning, and I’ll take everything that you said here into account for it! Do you have any requests or tips? I can implement just about anything. Want me to dm you a sample of a video that I’ve captioned a workflow that I made for this?

[-]