Paper Image

Hierarchical image classification with language models

Published on:

1 November 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Zhiyuan Ren,

Yiyang Su,

Xiaoming Liu

Bullets

Key Details

Constructs class hierarchies and tailored descriptions with language models

Compares classes at each level to generate distinguishing descriptions

Achieves state-of-the-art on 6 datasets without additional training

Provides interpretability into model decisions at each hierarchy level

Addresses CLIP's bias towards certain classes via comparisons

AI generated summary

Hierarchical image classification with language models

This paper proposes a new approach for zero-shot open-vocabulary image classification, which constructs hierarchies of class labels using language models like ChatGPT to generate comparative descriptions. The hierarchical structure and tailored descriptions for each level allow focusing on different distinguishing aspects of classes, overcoming limitations of vision-language models like CLIP.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up