Research map / Video / Video-language understanding
CNVid-3.5M: Build, Filter, and Pre-Train the Large-Scale Public Chinese Video-Text Dataset
Tian Gan, Qing Wang, Xingning Dong, Xiangyuan Ren, Liqiang Nie, Qingpei Guo
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 |
|---|---|---|---|
| pre-train on CNVid-3.5M rather than train without pre-training (ablation) | 2-E-F · VATEX (Chinese) | R@1: 36.4 -> 39.9 | Improved |
| filter weakly paired examples to the Top-3.5M split rather than use all 4.5M (ablation) | 2-E-F pre-trained on CNVid · MSRVTT (Translated) | R@1: 19.4 -> 20.7 | Improved |
| filter to the Top-3.0M split rather than Top-3.5M (ablation) | 2-E-F pre-trained on CNVid · VATEX (Chinese) | R@1: 39.9 -> 40.5 | Improved |
| filter to the Top-2.5M split rather than Top-3.5M (ablation) | 2-E-F pre-trained on CNVid · DiDemo (Translated) | R@1: 13.3 -> 12.1 | Hurt |
| add Hard Sample Curriculum Learning (ablation) | 2-E-F pre-trained on CNVid-3.5M · MSRVTT (Translated) | R@1: 20.7 -> 23.3 | Improved |
| add hard sample mining without curriculum learning (ablation) | 2-E-F pre-trained on CNVid-3.5M · VATEX (Chinese) | R@5: 77.2 -> 76.8 | Hurt |
| add HSCL (ablation) | 2-E-F pre-trained on COCO+VG+CC · MSRVTT | R@1: 28.6 -> 31.6 | Improved |
| add HSCL (ablation) | 2-E-F pre-trained on CC+WebVid · DiDemo | R@1: 29.0 -> 29.8 | Improved |
Closest papers by meaning
- Large Multilingual Models Pivot Zero-Shot Multimodal Learning across Languages (ICLR 2024 · similarity 0.86)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.86)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.86)
- InternVid: A Large-scale Video-Text Dataset for Multimodal Understanding and Generation (ICLR 2024 · similarity 0.85)
- Panda-70M: Captioning 70M Videos with Multiple Cross-Modality Teachers (CVPR 2024 · similarity 0.85)
- OpenVid-1M: A Large-Scale High-Quality Dataset for Text-to-video Generation (ICLR 2025 · similarity 0.85)
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.
