Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
I2MVFormer: Large Language Model Generated Multi-View Document Supervision for Zero-Shot Image Classification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 LLM-generated views and Wiki documents to I2MVFormer | I2MVFormer · AWA2, CUB, and FLO ZSL | T1: 76.4 -> 79.6; 45.4 -> 51.1; 40.0 -> 46.2 | Helped |
| Add LLM-generated views and Wiki documents to I2MVFormer | I2MVFormer · AWA2, CUB, and FLO GZSL | H: 71.5 -> 77.6; 43.8 -> 49.7; 51.5 -> 57.1 | Helped |
| Add SVSummary to global-only training (ablation) | I2MVFormer · AWA2, CUB, and FLO ZSL | T1: 73.6 -> 74.1; 45.6 -> 48.5; 38.9 -> 39.1 | Helped |
| Add local alignment loss to global alignment loss (ablation) | I2MVFormer · AWA2, CUB, and FLO ZSL | T1: 73.6 -> 78.4; 45.6 -> 49.0; 38.9 -> 43.2 | Helped |
| Add MVSummary to summarize per-view local tokens (ablation) | I2MVFormer with both losses · AWA2, CUB, and FLO ZSL | T1: 78.4 -> 79.6; 49.0 -> 51.1; 43.2 -> 46.2 | Helped |
| Use only local loss instead of global and local losses (ablation) | I2MVFormer · AWA2, CUB, and FLO ZSL | T1: 79.6 -> 57.7; 51.1 -> 32.5; 46.2 -> 24.2 | Hurt |
| Use unique 2-shot examples rather than repeating the same examples (ablation) | LLM view generation for I2MVFormer · AWA2 and FLO ZSL | T1: 73.1 -> 76.4; 43.1 -> 44.4 | Helped |
| Add examples to prompting rather than use zero-shot prompting (ablation) | LLM view generation for I2MVFormer · AWA2 and FLO ZSL | T1: 73.0 -> 74.2; 40.7 -> 42.1 | Helped |
| Use 3 LLM views rather than 1 LLM view (ablation) | I2MVFormer · AWA2 and FLO ZSL | T1: 71.6 -> 76.4; 39.0 -> 44.4 | Helped |
| Use 4 LLM views rather than 3 LLM views (ablation) | I2MVFormer · AWA2 and FLO ZSL | T1: 76.4 -> 76.6; 44.4 -> 44.5 | No clear effect |
Closest papers by meaning
- Improved Zero-Shot Classification by Adapting VLMs with Text Descriptions (CVPR 2024 · similarity 0.89)
- Liberating Seen Classes: Boosting Few-Shot and Zero-Shot Text Classification via Anchor Generation and Classification Reframing (AAAI 2024 · similarity 0.87)
- Epsilon: Exploring Comprehensive Visual-Semantic Projection for Multi-Label Zero-Shot Learning (AAAI 2025 · similarity 0.87)
- S3A: Towards Realistic Zero-Shot Classification via Self Structural Semantic Alignment (AAAI 2024 · similarity 0.87)
- Image2Sentence based Asymmetrical Zero-shot Composed Image Retrieval (ICLR 2024 · similarity 0.87)
- VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks (ICLR 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.