I indexed 669 GB of my GoPro videos using my M1 Max computer and local ML models

https://news.ycombinator.com/item?id=48222733 https://blog.simbastack.com/indexed-a-year-of-video-locally/

I wasn't familiar with your project though, interesting stuff.

I'm trying to add more photography related features to Framedex but yeah there's so much we can do locally, exciting times.

I'm really bullish on taking more video of my kids, with the thought that it will become easier and easier for AI to put them into little compilations I can enjoy later.

Aha, it makes total sense. This number sounds much more reasonable than “669 GB”, since the actual total size of processed frames would be like 10-30 GB.

(Not downplaying anything. Doing-at-home always requires some math on practicality)

> Total compute time 67h 40m 42s

I’m just curious tho — is there any paying options that can accelerate this kind of process? Just spin up GPU instances?

    M1 Max CPU is an ARM/SoC, comparable to an 11th gen Intel i9

ref: https://www.cpubenchmark.net/compare/4585vs4245/Apple-M1-Max...

EDITED: I didn't realize Whisper was a local model. I never tried transcription before, so I had always figured it was a pay model by OpenAI. I'll have to check it out (although the runtime listed here is a bit daunting).

For that project I'll say I don't see much degradation in embedding quality at much much worse quality than 720p (all the way down to 240p), which speeds things up considerably. Although I don't really do face or object detection, just scene embeddings. To me any process whereby it would take longer to process the video than watch it is probably a no go in general. Obviously a challenge for local-first analysis.

Take a fast, small and powerful LLM running locally to index my personal data like images, videos, documents and enrich them and tag with the enriched metadata.

Want to group by people - Search tagged metadata and group it What to search an image by description - tagged metadata What to organize by anything - tagged metadata

This should (hopefully) put an end to my file clutter

Being able to semantic search over your library is useful, but does it solve the review problem? I feel like you would still need to watch the footage back before you know what you're working with.

comes with some nifty features like NLE- integrations, people search, MCP, API etc

Disclaimer: one of the co-founders

The world and our discourse around it has changed so much over the past ten years and now with this kind of technology I'm so excited to be able to classify these images from my iCloud and start on the project.

Frame level embedding it covering a lot, but can miss out on a lot of action related searches.

I might be better off getting something with a beefy GPU on AWS or Google cloud.

When trying to read this article, the main website was throwing errors to CloudFlare unfortunately

Yep. I had the same problem.

> Then, run the frame analysis pipeline [...] I have a face recognition plugin using my custom faces data, object detection, on-screen text, shot type, and scene description [...] we will have three vector DB collections that have all the information about our videos, like video location metadata, camera name, faces recognized, objects detected, on-screen text, transcription, description of each scene, and many more [...] we can get better indexed data if you use the advanced mode indexing to use the Qwen2.5-VL-7B-Instruct model to understand and describe your video much better, but at a slower indexing speed

Yeah, uhm... ok :)

If anyone else has a similar problem, the real solution is as follows:

1. When recording, if you witness an interesting moment worth saving later, press the power button — this will mark the current moment in the video as a chapter.

2. Find the chapters later when editing and cut them into clips.

3. You're done :)

This has two main benefits over the insanity above:

1. It's trivially simple instead of insanely complex and inefficient.

2. It will reliably catch all the stuff you find interesting, since you're the one doing the marking.

The downsides:

1. Doesn't work retroactively.

2. It may miss interesting stuff if you miss it at the time as well.

3. Only works for this use case.

4. Nerds won't salivate over your usage of cutting edge tech.

Her client was recording while committing the abhorrent crime. The criminal would otherwise have got off.

From my perspective, the GoPro camera produced a good outcome. Still, one has wonder why anyone to record their criminal actions.