AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

ChatGPT-Powered Hierarchical Comparisons for Image Classification

Zhiyuan Ren, Yiyang Su, Xiaoming Liu

NeurIPS 2023 · Vision-language and multimodal › CLIP and zero-shot · approach: Adapting VLMs for zero-shot classification · 11 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace class-name CLIP prompts with hierarchical LLM comparative descriptionsCLIP ViT-L/14 · Describable Textures zero-shotaccuracy: 51.01 -> 58.04Improved
replace class-name CLIP prompts with hierarchical descriptionsCLIP ResNet-50 · ImageNet zero-shotaccuracy: 54.85 -> 60.63Improved
replace class-name OpenCLIP prompts with hierarchical descriptionsOpenCLIP ViT-L/14 · Describable Textures zero-shotaccuracy: 51.69 -> 61.79Improved
replace class-name OpenCLIP prompts with hierarchical descriptionsOpenCLIP · Food101 zero-shotaccuracy: 93.41 -> 92.68Hurt
replace class-name CLIP prompts with hierarchical descriptionsCLIP ViT-L/14 · CUB zero-shotaccuracy: 63.46 -> 65.45Improved
replace non-comparative LLM descriptions (classification by description) with hierarchical comparative descriptionsCLIP ViT-L/14 · ImageNet zero-shotaccuracy: 75.36 -> 79.64Improved
replace non-comparative LLM descriptions with hierarchical comparative descriptionsCLIP ViT-L/14 · mean over six datasetsaccuracy: 70.26 -> 71.71Improved
set score momentum lambda to 0 (no offset) (ablation)hierarchical scoring · mean over six datasets, ViT-B/32accuracyHurt
use larger group number N (fewer hierarchy levels, ~sixfold speedup) (ablation)knowledge tree · ImageNet ViT-B/32accuracy: 1.64% dipHurt
increase depth of descriptions used in prediction (ablation)hierarchical scoring · ImageNet ViT-B/32accuracyImproved
vary score reduction tolerance tau from 0 to 1 (ablation)hierarchical scoring · mean over six datasets, ViT-B/32accuracyNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy