Research map / Vision-language and multimodal / CLIP and zero-shot
I2DFormer: Learning Image to Document Attention for Zero-Shot Image Classification
Muhammad Ferjad Naeem, Yongqin Xian, Luc Van Gool, Federico Tombari
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 I2D Attention to I2DGlobal (ablation) | I2DFormer · AWA2, CUB, and FLO | ZSL T1: 69.4 -> 76.4; 37.2 -> 45.4; 37.2 -> 40.0 | Improved |
| replace I2D Attention with FILIP (ablation) | I2DGlobal · AWA2, CUB, and FLO | ZSL T1: 69.4 -> 67.3; 37.2 -> 35.7; 37.2 -> 38.3 | Hurt |
| replace I2D Attention with ViLBERT cross-modal and self-attention (ablation) | I2DGlobal · AWA2, CUB, and FLO | ZSL T1: 69.4 -> 75.0; 37.2 -> 29.9; 37.2 -> 21.3 | Mixed |
| use LongFormer instead of GloVe input embeddings (ablation) | I2DFormer · AWA2, CUB, and FLO | ZSL T1: 76.4 -> 51.5; 45.4 -> 34.3; 40.0 -> 21.8 | Hurt |
| use MPNet instead of GloVe input embeddings (ablation) | I2DFormer · AWA2, CUB, and FLO | ZSL T1: 76.4 -> 65.5; 45.4 -> 36.1; 40.0 -> 28.3 | Hurt |
| use Word2Vec instead of GloVe input embeddings (ablation) | I2DFormer · AWA2, CUB, and FLO | ZSL T1: 76.4 -> 74.0; 45.4 -> 43.8; 40.0 -> 37.8 | Hurt |
| learn image-document semantic embeddings from documents rather than class-name GloVe embeddings | I2DFormer global module · AWA2, CUB, and FLO | ZSL T1: 52.1 -> 61.6; 20.4 -> 29.0; 21.6 -> 25.8 | Improved |
| use I2DEmb instead of GloVe document embeddings | semantic embedding methods · AWA2, CUB, and FLO | ZSL T1: 61.6 -> 76.4; 29.0 -> 45.4; 25.8 -> 40.0 | Improved |
| use I2DFormer rather than the strongest unsupervised embedding baseline | I2DFormer · AWA2, CUB, and FLO | GZSL H: 66.0 -> 71.5; 37.3 -> 43.8; 44.3 -> 51.5 | Improved |
Later papers that compared with this method
1 later accepted paper reports I2DFormer+ as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- ZeroMamba: Exploring Visual State Space Model for Zero-Shot Learning (AAAI 2025 · better on 1 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- I2DFormer+: Learning Image to Document Summary Attention for Zero-Shot Image Classification (IJCV 2024 · similarity 0.99)
- I2MVFormer: Large Language Model Generated Multi-View Document Supervision for Zero-Shot Image Classification (CVPR 2023 · similarity 0.92)
- VGSE: Visually-Grounded Semantic Embeddings for Zero-Shot Learning (CVPR 2022 · similarity 0.90)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.89)
- Context-I2W: Mapping Images to Context-Dependent Words for Accurate Zero-Shot Composed Image Retrieval (AAAI 2024 · similarity 0.88)
- Progressive Semantic-Guided Vision Transformer for Zero-Shot Learning (CVPR 2024 · 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.
