this post was submitted on 12 Mar 2025
6 points (100.0% liked)

LocalLLaMA

2950 readers
13 users here now

Welcome to LocalLLaMA! Here we discuss running and developing machine learning models at home. Lets explore cutting edge open source neural network technology together.

Get support from the community! Ask questions, share prompts, discuss benchmarks, get hyped at the latest and greatest model releases! Enjoy talking about our awesome hobby.

As ambassadors of the self-hosting machine learning community, we strive to support each other and share our enthusiasm in a positive constructive way.

founded 2 years ago
MODERATORS
 

I'd like something to describe images for me and also recognise any text contained in them. I've tried llama3. 2-vision, llava and minicpm-v but they all get the text recognition laughably wrong.

Or maybe I should lay my image recognition dreams to rest with my measly 8 GB RAM card.

Edit: gemma3:4b is even worse than the others. It doesn't even find the text and hallucinates others.

you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 3 points 1 month ago (1 children)

Perhaps a multi stage approach is a better idea? Find a model that's exclusively for optical character recognition, then feed it's output into an llm

[–] [email protected] 2 points 1 month ago (1 children)

Why didn't I think of this? granite3.2-vision seems to do the trick, even though it doesn't know the German letter ß. But I can live without that.

Thanks!

[–] [email protected] 1 points 1 month ago* (last edited 1 month ago)

I can't you why you didn't. But I can tell you a trick to come up with better ideas: force yourself to come up with 10 ideas, no matter how stupid. Then only filter them once you're done.

This causes you to take less mental shortcuts. And you end up with better results (usually) at the cost of more mental effort/energy