Research map / Vision-language and multimodal / CLIP and zero-shot
CyCLIP: Cyclic Contrastive Language-Image Pretraining
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 cross-modal and in-modal cycle consistency regularizers | CLIP RN50 on CC3M · ImageNet1K zero-shot | Top-1 accuracy: 20.03 -> 22.08 | Helped |
| add cycle consistency regularizers | CLIP RN50 on CC3M · CIFAR-100 zero-shot | Top-1 accuracy: 18.69 -> 23.15 | Helped |
| add cycle consistency regularizers | CLIP RN50 on CC3M · CIFAR-10 zero-shot | Top-1 accuracy: 46.54 -> 51.45 | Helped |
| add cycle consistency regularizers | CLIP RN50 on CC3M · ImageNet-A zero-shot | Top-1 accuracy: 4.23 -> 5.35 | Helped |
| add cycle consistency regularizers | CLIP · CIFAR-10 consistency score | Top-1 consistency: 44.60 -> 48.81 | Helped |
| use only cross-modal consistency (drop in-modal) (ablation) | CyCLIP · CIFAR-10 zero-shot | Top-1 accuracy: 51.45 -> 55.52 | Helped |
| use only cross-modal consistency (drop in-modal) (ablation) | CyCLIP · ImageNet1K zero-shot | Top-1 accuracy: 22.08 -> 21.73 | Hurt |
| use only in-modal consistency (drop cross-modal) (ablation) | CyCLIP · ImageNet1K zero-shot | Top-1 accuracy: 22.08 -> 21.21 | Hurt |
| add cycle consistency regularizers | CLIP · 14 linear probe datasets | accuracy: better on 9 of 14 | Mixed |
| add cycle consistency regularizers | CLIP · Flickr30K and MSCOCO zero-shot retrieval | R@1: 30.9 vs 29.9 image R@1 Flickr | No clear effect |
| add cycle consistency regularizers | CLIP · Flickr30K and MSCOCO fine-tuned image retrieval | R@1: 46.3 -> 47.3 | Helped |
| add cycle consistency regularizers (varying data size 500K-4M) | CLIP · ImageNet1K | accuracy | Helped |
Closest papers by meaning
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.91)
- FineCLIP: Self-distilled Region-based CLIP for Better Fine-grained Understanding (NeurIPS 2024 · similarity 0.91)
- MLIP: Efficient Multi-Perspective Language-Image Pretraining with Exhaustive Data Utilization (ICML 2024 · similarity 0.90)
- SuperCLIP: CLIP with Simple Classification Supervision (NeurIPS 2025 · similarity 0.90)
- TripletCLIP: Improving Compositional Reasoning of CLIP via Synthetic Vision-Language Negatives (NeurIPS 2024 · similarity 0.89)
- Improving fine-grained understanding in image-text pre-training (ICML 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.