Research map / Vision-language and multimodal / CLIP and zero-shot
CALIP: Zero-Shot Enhancement of CLIP with Parameter-Free Attention
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 parameter-free bidirectional cross-modal attention | CLIP · ImageNet | accuracy: 83.94% -> 87.51% | Helped |
| Add parameter-free bidirectional cross-modal attention | CLIP · ImageNet-V2, ImageNet-A, ImageNet-R, ImageNet-Sketch | accuracy: 53.27 -> 53.70; 23.61 -> 23.96; 60.42 -> 60.81; 35.44 -> 35.61 | Helped |
| Add parameter-free bidirectional cross-modal attention | CLIP · ISIC | accuracy: +18.15% | Helped |
| Add parameter-free bidirectional cross-modal attention | CLIP · BCCD | accuracy: +13.76% | Helped |
| Add learnable cross-modal projections and 16-shot fine-tuning | CLIP · ImageNet | accuracy: 91.83% -> 94.20% | Helped |
| Combine original logits with visual-guided logits (ablation) | CALIP · Caltech101 | accuracy: 83.94% -> 84.02% | Helped |
| Combine original logits with textual-blended logits (ablation) | CALIP · Caltech101 | accuracy: 83.94% -> 85.10% | Helped |
| Add logits formed from both attention-updated modalities (ablation) | CALIP · Caltech101 | accuracy: 85.66% -> 85.34% | Hurt |
| Add logits formed from both attention-updated modalities (ablation) | CALIP-FS · Caltech101 | accuracy: 94.75% -> 94.66% | Hurt |
| Add pre-projection layers for visual and textual features (ablation) | CALIP-FS · Caltech101 | accuracy: 87.71% -> 89.75% | Helped |
| Add textual post-projection while retaining both pre-projections (ablation) | CALIP-FS · Caltech101 | accuracy: 89.75% -> 90.36% | Helped |
| Add visual post-projection while retaining both pre-projections (ablation) | CALIP-FS · Caltech101 | accuracy: 89.75% -> 93.94% | Helped |
Closest papers by meaning
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.91)
- Understanding Transferable Representation Learning and Zero-shot Transfer in CLIP (ICLR 2024 · similarity 0.91)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.91)
- SuperCLIP: CLIP with Simple Classification Supervision (NeurIPS 2025 · similarity 0.91)
- un$^2$CLIP: Improving CLIP's Visual Detail Capturing Ability via Inverting unCLIP (NeurIPS 2025 · similarity 0.90)
- Demystifying CLIP Data (ICLR 2024 · similarity 0.90)
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.