AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

Textual Knowledge Matters: Cross-Modality Co-Teaching for Generalized Visual Class Discovery

Haiyang Zheng, Nan Pu, Wenjing Li, Nicu Sebe, Zhun Zhong

ECCV 2024 · Vision-language and multimodal › Cross-modal learning · approach: Probabilistic embeddings for cross-modal retrieval · 15 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
add cross-modality co-teaching (ablation)TextGCD baseline image classifier · CUBAll accuracy: 49.0 -> 71.7Improved
add image-text contrastive learning (ablation)TextGCD · CUBAll accuracy (image classifier): 71.7 -> 73.9Improved
add soft voting over both modalities (ablation)TextGCD · CUBAll accuracy: 74.8 -> 76.6Improved
add class-aligning stage (text guides image) (ablation)TextGCD · ImageNet-100New accuracy: 76.0 -> 85.2Improved
add class-aligning stage (text guides image) (ablation)TextGCD · CIFAR-100New accuracy: 74.1 -> 84.6Improved
add warm-up stage (ablation)TextGCD · CUBAll accuracy: 74.8 -> 76.5Improved
add warm-up stage (ablation)TextGCD · CIFAR-100All accuracy: 83.3 -> 82.8Hurt
guide text model with image model (I->T) instead of text guiding image (ablation)TextGCD class aligning · CIFAR-100All accuracy: 85.7 -> 84.6Hurt
use cross-modality co-teaching instead of single-modality image co-teaching (ablation)TextGCD · CUBAll accuracy: 60.3 -> 76.6Improved
increase text tags from 1 to 3 plus 2 attributes (ablation)RTG text description · CUBAll accuracy: 67.9 -> 76.6Improved
use ViT-B/16 CLIP instead of ViT-H-14 for TextGCD training backbone (ablation)TextGCD · CIFAR-100All accuracy: 85.7 vs 86.4No clear effect
use stronger auxiliary model FLIP instead of CLIP for text generation (ablation)RTG · CUBAll accuracy: 76.6 -> 79.5Improved

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