Research map / Vision-language and multimodal / Human and embodied modeling
Multimodal C4: An Open, Billion-scale Corpus of Images Interleaved with Text
Wanrong Zhu, Jack Hessel, Anas Awadalla, Samir Yitzhak Gadre, Jesse Dodge, Alex Fang, Youngjae Yu, Ludwig Schmidt, William Yang Wang, Yejin Choi
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 |
|---|---|---|---|
| Use zero-shot CLIP ViT-L/14 for image-sentence matching | Within-document image-text alignment · Story-DII | AUC: 92.8 vs. 85.5 | Improved |
| Use zero-shot CLIP ViT-L/14 for image-sentence matching | Within-document image-text alignment · Story-SIS | P@1: 73.3 vs. 53.1 | Improved |
| Use zero-shot CLIP ViT-L/14 for image-sentence matching | Within-document image-text alignment · DII-Stress | AUC: 98.7 vs. 95.3 | Improved |
| Use zero-shot CLIP ViT-L/14 for image-sentence matching | Within-document image-text alignment · RQA | AUC: 80.7 vs. 69.3 | Improved |
| Use zero-shot CLIP ViT-L/14 for image-sentence matching | Within-document image-text alignment · DIY | AUC: 74.0 vs. 61.8 | Improved |
| Replace max-similarity matching with bipartite linear assignment (ablation) | Within-document image-text assignment · mmc4 document sequences | Sentences associated with an image: 22% -> 34% | Improved |
| Replace max-similarity matching with bipartite linear assignment (ablation) | Within-document image-text assignment · mmc4 document sequences | Mean CLIP similarity: 24.5 -> 24.0 | Hurt |
| Train on interleaved mmc4-core sequences instead of image-caption pairs | OpenFlamingo-3B · COCO image captioning, 4-shot and 8-shot | CIDEr: 20-30 CIDEr point improvements | Improved |
| Train on image-caption pairs rather than interleaved sequences (ablation) | OpenFlamingo-3B · COCO image captioning, 4-shot to 8-shot | CIDEr | Hurt |
Closest papers by meaning
- OmniCorpus: A Unified Multimodal Corpus of 10 Billion-Level Images Interleaved with Text (ICLR 2025 · similarity 0.90)
- CoMM: A Coherent Interleaved Image-Text Dataset for Multimodal Understanding and Generation (CVPR 2025 · similarity 0.88)
- A High Quality Dataset and Reliable Evaluation for Interleaved Image-Text Generation (ICLR 2026 · similarity 0.88)
- OpenING: A Comprehensive Benchmark for Judging Open-ended Interleaved Image-Text Generation (CVPR 2025 · similarity 0.88)
- Learning Interleaved Image-Text Comprehension in Vision-Language Large Models (ICLR 2025 · similarity 0.87)
- CapsFusion: Rethinking Image-Text Data at Scale (CVPR 2024 · 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.
