Research map / Vision-language and multimodal / Cross-modal learning
COOKIE: Contrastive Cross-Modal Knowledge Sharing Pre-Training for Vision-Language Representation
Keyu Wen, Jin Xia, Yuanyuan Huang, Linyang Li, Jiayan Xu, Jie Shao
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 cross-modal contrastive pre-training (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 526.9 -> 545.4 | Improved |
| add visual contrastive learning on top of CCL (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 545.4 -> 547.8 | Improved |
| add textual contrastive learning on top of CCL (ablation) | COOKIE · STS-B text matching | Pearson/Spearman mean: 72.4 -> 75.1 | Improved |
| add visual contrastive learning on top of CCL (ablation) | COOKIE · STS-B text matching | Pearson/Spearman mean: 72.4 -> 72.1 | No clear effect |
| add weight-sharing transformer encoder (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 536.6 -> 547.5 | Improved |
| share weights in WS-TE between image and text (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 541.6 -> 547.5 | Improved |
| add text-aligned visual transformer (TAV-TE) with weight sharing (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 547.5 -> 550.0 | Improved |
| replace WS-TE with FC layers (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 536.6 -> 536.8 | No clear effect |
| use 2 WS-TE layers instead of 1 with pre-training (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 547.8 -> 550.0 | Improved |
| use 8 WS-TE layers instead of 2 without pre-training (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 526.9 -> 250.8 | Hurt |
| increase pre-training corpus from CC 2.8M to 5.9M pairs (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 540.5 -> 550.0 | Improved |
| add VQA and GQA to pre-training corpus (ablation) | COOKIE · MSCOCO 1K test ITM | Rsum: 547.6 -> 550.0 | 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.
- Uniter-large — UNITER: UNiversal Image-TExt Representation Learning (ECCV 2020) · better on 1 of 1 shared result
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
- Vision-Language Pre-Training With Triple Contrastive Learning (CVPR 2022 · similarity 0.90)
- Dynamic Adapter with Semantics Disentangling for Cross-lingual Cross-modal Retrieval (AAAI 2025 · similarity 0.88)
- CL2CM: Improving Cross-Lingual Cross-Modal Retrieval via Cross-Lingual Knowledge Transfer (AAAI 2024 · similarity 0.88)
- Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data (ICLR 2024 · similarity 0.88)
- WikiDO: A New Benchmark Evaluating Cross-Modal Retrieval for Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Learning Shared Representations from Unpaired Data (NeurIPS 2025 · similarity 0.87)
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.
