Research map / Video / Video generation
TIP-I2V: A Million-Scale Real Text and Image Prompt Dataset for Image-to-Video Generation
Wenhao Wang, Yi Yang
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 |
|---|---|---|---|
| train a VideoMAE video classifier instead of using frame-level fake image detectors | real-vs-generated video detection · TIP-I2V detection | accuracy: 53.1 -> 96.3 | Improved |
| evaluate detector cross-domain instead of same-domain | VideoMAE classifier · TIP-I2V three-way classification | accuracy: 96.3 -> 87.2 | Hurt |
| evaluate detector cross-domain instead of same-domain | VideoMAE classifier · TIP-I2V three-way classification | mAP: 99.4 -> 94.8 | Hurt |
| train a CosFace metric-learning model for source tracing instead of pre-trained image copy detection models | SSCD · TIP-Trace | muAP: 77.5 -> 84.7 | Improved |
| test on I2VGen-XL cross-domain instead of same-domain | CosFace baseline · TIP-Trace | muAP: -17.0% muAP | Hurt |
| use real-user prompts (TIP-Eval) instead of expert-designed prompts | Pika vs CogVideoX-5B · TIP-Eval | 10 VBench dimensions: 8 out of 10 | Mixed |
| use CogVideoX-5B on calligraphy subject instead of cottage | CogVideoX-5B · TIP-Eval | aesthetic quality: 0.74 -> 0.40 | Hurt |
| use middle frame with fake image detectors on image-generated videos | UnivFD and other detectors · TIP-I2V detection | accuracy: 53.1 | No clear effect |
| compare existing image copy detection models on tracing | SSCD · TIP-Trace | muAP: 77.5 | No clear effect |
| compare video-text alignment across I2V models | all five I2V models · TIP-Eval | video-text alignment: 0.26 | No clear effect |
Closest papers by meaning
- VidProM: A Million-scale Real Prompt-Gallery Dataset for Text-to-Video Diffusion Models (NeurIPS 2024 · similarity 0.93)
- VideoUFO: A Million-Scale User-Focused Dataset for Text-to-Video Generation (NeurIPS 2025 · similarity 0.91)
- OpenVid-1M: A Large-Scale High-Quality Dataset for Text-to-video Generation (ICLR 2025 · similarity 0.90)
- VideoBooth: Diffusion-based Video Generation with Image Prompts (CVPR 2024 · similarity 0.90)
- A Recipe for Scaling up Text-to-Video Generation with Text-free Videos (CVPR 2024 · similarity 0.90)
- A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis (AAAI 2024 · 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.
