3 ms·
So 10 years ago this author manually examined all the photos and tagged them to find the conditional probabilities and all the neat stats-blog content. I'd lov
by floatrock 3y ago
So 10 years ago this author manually examined all the photos and tagged them to find the conditional probabilities and all the neat stats-blog content.
I'd love to see a redo of this post using today's AI tools. Like, I know midjourney can spit out images "in the style of bob ross", but I mean how do you use these tools to analyze features across the bob ross corpus and ask "How many pictures contain happy little clouds?" And can these tools answer more abstract questions like "what kinds of bob ross paintings tend to have happy little clouds vs paintings with no clouds?"
- brucethemoose2 3y agoOh yes we can! Its already a component of these generative tools: https://huggingface.co/spaces/pharma/CLIP-Interrogator https://huggingface.co/spaces/pharma/CLIP-Interrogator You might want to finetune it to reliably recognize "happy little clouds" with some samples of what you consider to be happy little clouds, as well as not happy and not little clouds for regularization.
- brucethemoose2 3y agoAlso, see LLaVA, aka CLIP hooked up to an LLM, if you are looking for more verbose and logical analysis of images: https://llava-vl.github.io/ https://llava-vl.github.io/