Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

>> It's also 100% private. Make any media searchable without it ever leaving your computer.

It seems you are using the CLIP model, which you can run on the CPU. Would you have any estimates of how long the indexing would take? Also curious how often you sample videos (every minute? every 10sec?)

One of the troubles I have is the large library of family videos, which become especially difficult to index. I've been thinking of everything from sample-indexing to scene detection etc.



All the images get scaled down before the app computes the embeddings, so it should only take a few minutes, but definitely depends on how much you're trying to index at once.

I'm making improvements to make indexing more resilient and faster. The videos are sampled once every second. That's something I'm tweaking for better performance as well. Was considering letting the user adjust this too.

How have you tried indexing your videos so far?


Yeah, buy a 3090, put it in a Ubuntu closet box, set it up as a server, and let it rip. It’s really not hard or expensive.

I’m tired of people trying to force a square peg (MacBooks) into a round hole (ml compute). Apple needs to get on board with making it easier and more cost effective for people to access compute more effectively. Developers are jumping through hoops to accommodate apples business goals of locking you to their ecosystem.


Sounds cool! I haven't explored that but there are lots of interesting ways to approach this type of thing. Tons of creative ways to use these models.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: