Research map / Vision-language and multimodal / CLIP and zero-shot
RA-CLIP: Retrieval Augmented Contrastive Language-Image Pre-Training
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 |
|---|---|---|---|
| augment image embeddings with retrieved reference image-text pairs | CLIP trained on YFCC15M · ImageNet zero-shot | top-1 accuracy: 37.7 -> 53.5 | Helped |
| augment image embeddings with retrieved reference image-text pairs | CLIP · 10 image-classification datasets, zero-shot average | accuracy: 39.3 -> 52.0 | Helped |
| augment image embeddings with retrieved reference image-text pairs | CLIP · 10 image-classification datasets, linear-probe average | accuracy: 68.6 -> 75.5 | Helped |
| augment image embeddings with retrieved reference image-text pairs | RegionCLIP · COCO zero-shot ROI | AP: 44.4 -> 48.4 | Helped |
| augment image embeddings with retrieved reference image-text pairs | RegionCLIP · LVIS zero-shot ROI | AP: 21.6 -> 23.2 | Helped |
| use reference pairs from CC instead of YFCC (ablation) | RA-CLIP · ImageNet zero-shot | top-1 accuracy: 53.5 -> 54.5 | Helped |
| use reference pairs from LAION instead of YFCC (ablation) | RA-CLIP · ImageNet zero-shot | top-1 accuracy: 53.5 -> 54.2 | Helped |
| replace the frozen SentenceT reference text encoder with RA-CLIP's trained text encoder (ablation) | RA-CLIP · ImageNet zero-shot | top-1 accuracy: 54.5 -> 54.4 | No clear effect |
| initialize the CLIP image encoder with ImageNet-pretrained ViT (ablation) | CLIP · ImageNet zero-shot | top-1 accuracy: 37.7 -> 46.1 | Helped |
| augment the text representation as well as the image representation (ablation) | RA-CLIP · ImageNet zero-shot | top-1 accuracy: 54.5 -> 53.1 | Hurt |
| retain the original image embedding and fuse it with both augmented embeddings (ablation) | RAM · ImageNet zero-shot | top-1 accuracy: 51.8 -> 54.5 | Helped |
| retrieve K=16 pairs instead of K=64 (ablation) | RA-CLIP · ImageNet zero-shot | top-1 accuracy: 54.5 -> 54.3 | No clear effect |
Closest papers by meaning
- Meta CLIP 2: A Worldwide Scaling Recipe (NeurIPS 2025 · similarity 0.89)
- Demystifying CLIP Data (ICLR 2024 · similarity 0.89)
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.89)
- CLIP-KD: An Empirical Study of CLIP Model Distillation (CVPR 2024 · similarity 0.89)
- TripletCLIP: Improving Compositional Reasoning of CLIP via Synthetic Vision-Language Negatives (NeurIPS 2024 · similarity 0.88)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.88)
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.