Research map / Vision-language and multimodal / CLIP and zero-shot
RANKCLIP: Ranking-Consistent Language-Image Pretraining
Yiming Zhang, Zhuokai Zhao, Zhaorun Chen, Zhili Feng, Zenghui Ding, Yining Sun
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 list-wise ranking consistency loss to CLIP | CLIP ViT-B/32 CC3M · ImageNet1K zero-shot | Top-1 accuracy: 9.06% -> 10.16% | Improved |
| add ranking consistency loss with RN50 backbone | CLIP RN50 · ImageNet1K zero-shot | Top-1 accuracy: 21.6% -> 30.9% | Improved |
| add ranking consistency loss with ViT-B/32 backbone | CLIP ViT-B/32 · ImageNet1K zero-shot | Top-1 accuracy: 20.7% -> 26.2% | Improved |
| add ranking consistency loss | CLIP RN50 · Linear probing average | accuracy: 64.2% -> 68.9% | Improved |
| add ranking consistency loss | CLIP ViT-B/32 · Linear probing average | accuracy: 60.7% -> 61.3% | Improved |
| add ranking consistency loss | CLIP ViT-B/32 CC3M · MSCOCO retrieval Recall@10 | Recall@10: image-to-text 26.94% -> 28.48%; text-to-image 14.04% -> 14.18% | Mixed |
| add ranking consistency loss | CLIP ViT-B/32 CC3M · ImageNet-R | Top-1 accuracy: 9.36% -> 11.34% | Improved |
| add ranking consistency loss | CLIP ViT-B/32 CC3M · SVHN linear probing | accuracy: 50.5% -> 47.7% | Hurt |
| use cross-modal rank loss only (drop in-modal) (ablation) | RankCLIP · ImageNet1K alignment/uniformity | uniformity score S_U: -0.13 -> -0.14 | Mixed |
| use in-modal rank loss only (drop cross-modal) (ablation) | RankCLIP · ImageNet1K alignment/uniformity | uniformity score S_U: -0.13 -> -0.17 | Hurt |
| add ranking consistency loss | CLIP · ImageNet1K alignment | alignment score S_A: 0.33 -> 0.29 | Hurt |
| use adaptive lambda schedule instead of fixed weights (ablation) | RankCLIP · ImageNet1K and MSCOCO | accuracy and recall | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CyCLIP — CyCLIP: Cyclic Contrastive Language-Image Pretraining (NeurIPS 2022) · better on 4 of 4 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
- SuperCLIP: CLIP with Simple Classification Supervision (NeurIPS 2025 · similarity 0.91)
- Demystifying CLIP Data (ICLR 2024 · similarity 0.91)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.91)
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 2025 · similarity 0.90)
- TripletCLIP: Improving Compositional Reasoning of CLIP via Synthetic Vision-Language Negatives (NeurIPS 2024 · similarity 0.90)
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · 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.
