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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add cross-modality co-teaching (ablation) | TextGCD baseline image classifier · CUB | All accuracy: 49.0 -> 71.7 | Improved |
| add image-text contrastive learning (ablation) | TextGCD · CUB | All accuracy (image classifier): 71.7 -> 73.9 | Improved |
| add soft voting over both modalities (ablation) | TextGCD · CUB | All accuracy: 74.8 -> 76.6 | Improved |
| add class-aligning stage (text guides image) (ablation) | TextGCD · ImageNet-100 | New accuracy: 76.0 -> 85.2 | Improved |
| add class-aligning stage (text guides image) (ablation) | TextGCD · CIFAR-100 | New accuracy: 74.1 -> 84.6 | Improved |
| add warm-up stage (ablation) | TextGCD · CUB | All accuracy: 74.8 -> 76.5 | Improved |
| add warm-up stage (ablation) | TextGCD · CIFAR-100 | All accuracy: 83.3 -> 82.8 | Hurt |
| guide text model with image model (I->T) instead of text guiding image (ablation) | TextGCD class aligning · CIFAR-100 | All accuracy: 85.7 -> 84.6 | Hurt |
| use cross-modality co-teaching instead of single-modality image co-teaching (ablation) | TextGCD · CUB | All accuracy: 60.3 -> 76.6 | Improved |
| increase text tags from 1 to 3 plus 2 attributes (ablation) | RTG text description · CUB | All accuracy: 67.9 -> 76.6 | Improved |
| use ViT-B/16 CLIP instead of ViT-H-14 for TextGCD training backbone (ablation) | TextGCD · CIFAR-100 | All accuracy: 85.7 vs 86.4 | No clear effect |
| use stronger auxiliary model FLIP instead of CLIP for text generation (ablation) | RTG · CUB | All accuracy: 76.6 -> 79.5 | Improved |
Closest papers by meaning
- GET: Unlocking the Multi-modal Potential of CLIP for Generalized Category Discovery (CVPR 2025 · similarity 0.91)
- Semantic-Guided Novel Category Discovery (AAAI 2024 · similarity 0.88)
- HiLo: A Learning Framework for Generalized Category Discovery Robust to Domain Shifts (ICLR 2025 · similarity 0.88)
- Less Attention is More: Prompt Transformer for Generalized Category Discovery (CVPR 2025 · similarity 0.87)
- SPTNet: An Efficient Alternative Framework for Generalized Category Discovery with Spatial Prompt Tuning (ICLR 2024 · similarity 0.87)
- SEAL: Semantic-Aware Hierarchical Learning for Generalized Category Discovery (NeurIPS 2025 · similarity 0.87)
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.
