Research map / Vision-language and multimodal / CLIP and zero-shot
I2DFormer+: Learning Image to Document Summary 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 I2D Global (ablation) | I2DFormer · AWA2 (ZSL) | top-1 accuracy: 69.4 -> 76.4 | Improved |
| add I2D Attention to I2D Global (ablation) | I2DFormer · CUB (ZSL) | top-1 accuracy: 37.2 -> 45.4 | Improved |
| add I2D Attention to I2D Global (ablation) | I2DFormer · FLO (ZSL) | top-1 accuracy: 37.2 -> 40.0 | Improved |
| add FILIP patch-to-word attention to I2D Global (ablation) | I2DGlobal · AWA2, CUB, FLO (ZSL) | top-1 accuracy: 69.4 -> 67.3 (AWA2), 37.2 -> 35.7 (CUB), 37.2 -> 38.3 (FLO) | Mixed |
| add ViLBERT cross-modal attention to I2D Global (ablation) | I2DGlobal · AWA2, CUB, FLO (ZSL) | top-1 accuracy: 69.4 -> 75.0 (AWA2), 37.2 -> 29.9 (CUB), 37.2 -> 21.3 (FLO) | Mixed |
| use LongFormer instead of GloVe as document input (ablation) | I2DFormer · CUB (ZSL) | top-1 accuracy: 45.4 -> 34.3 | Hurt |
| use MPNet instead of GloVe as document input (ablation) | I2DFormer · CUB (ZSL) | top-1 accuracy: 45.4 -> 36.1 | Hurt |
| use Word2Vec instead of GloVe as document input (ablation) | I2DFormer · CUB (ZSL) | top-1 accuracy: 45.4 -> 43.8 | Hurt |
| use documents instead of class names for GloVe embedding | I2D Global · CUB (ZSL) | top-1 accuracy: 20.4 -> 29.0 | Improved |
| learn embedding with I2DFormer instead of averaged GloVe document | I2D Global · FLO (ZSL) | top-1 accuracy: 25.8 -> 40.0 | Improved |
| replace GloVe with learned I2DEmb | f-VAEGAN-D2 · AWA2 (ZSL) | top-1 accuracy: 70.7 -> 85.1 | Improved |
| replace GloVe with learned I2DEmb | GAZSL · AWA2 (ZSL) | top-1 accuracy: 63.7 -> 83.1 | Improved |
Closest papers by meaning
- I2DFormer: Learning Image to Document Attention for Zero-Shot Image Classification (NeurIPS 2022 · similarity 0.99)
- 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 0.87)
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.87)
- Progressive Semantic-Guided Vision Transformer for Zero-Shot Learning (CVPR 2024 · similarity 0.86)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.86)
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.
