james-oldfield/PoS-subspaces

[NeurIPS'23] Parts of Speech–Grounded Subspaces in Vision-Language Models

21
/ 100
Experimental

This project helps researchers and developers working with Vision-Language Models like CLIP to better understand and control how these models interpret images. It takes image representations and associated text descriptions, then separates the visual information into distinct components based on parts of speech (e.g., nouns for objects, adjectives for appearance). This allows users to extract or manipulate specific visual attributes more precisely.

No commits in the last 6 months.

Use this if you are a researcher or developer who needs to disentangle different visual aspects (like object vs. style) within your vision-language model embeddings for tasks like controlled image generation or improved classification.

Not ideal if you are looking for an off-the-shelf application to directly edit images or perform general-purpose image classification without delving into model embeddings.

vision-language modeling image representation analysis generative AI control computer vision research model interpretability
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 7 / 25
Maturity 8 / 25
Community 6 / 25

How are scores calculated?

Stars

29

Forks

2

Language

Jupyter Notebook

License

Last pushed

Feb 25, 2024

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/james-oldfield/PoS-subspaces"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.