Phantom-Data : Towards a General Subject-Consistent Video Generation Dataset
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915356178644992 |
|---|---|
| author | Chen, Zhuowei Li, Bingchuan Ma, Tianxiang Liu, Lijie Liu, Mingcong Zhang, Yi Li, Gen Li, Xinghui Zhou, Siyu He, Qian Wu, Xinglong |
| author_facet | Chen, Zhuowei Li, Bingchuan Ma, Tianxiang Liu, Lijie Liu, Mingcong Zhang, Yi Li, Gen Li, Xinghui Zhou, Siyu He, Qian Wu, Xinglong |
| contents | Subject-to-video generation has witnessed substantial progress in recent years. However, existing models still face significant challenges in faithfully following textual instructions. This limitation, commonly known as the copy-paste problem, arises from the widely used in-pair training paradigm. This approach inherently entangles subject identity with background and contextual attributes by sampling reference images from the same scene as the target video. To address this issue, we introduce \textbf{Phantom-Data, the first general-purpose cross-pair subject-to-video consistency dataset}, containing approximately one million identity-consistent pairs across diverse categories. Our dataset is constructed via a three-stage pipeline: (1) a general and input-aligned subject detection module, (2) large-scale cross-context subject retrieval from more than 53 million videos and 3 billion images, and (3) prior-guided identity verification to ensure visual consistency under contextual variation. Comprehensive experiments show that training with Phantom-Data significantly improves prompt alignment and visual quality while preserving identity consistency on par with in-pair baselines. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_18851 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Phantom-Data : Towards a General Subject-Consistent Video Generation Dataset Chen, Zhuowei Li, Bingchuan Ma, Tianxiang Liu, Lijie Liu, Mingcong Zhang, Yi Li, Gen Li, Xinghui Zhou, Siyu He, Qian Wu, Xinglong Computer Vision and Pattern Recognition Subject-to-video generation has witnessed substantial progress in recent years. However, existing models still face significant challenges in faithfully following textual instructions. This limitation, commonly known as the copy-paste problem, arises from the widely used in-pair training paradigm. This approach inherently entangles subject identity with background and contextual attributes by sampling reference images from the same scene as the target video. To address this issue, we introduce \textbf{Phantom-Data, the first general-purpose cross-pair subject-to-video consistency dataset}, containing approximately one million identity-consistent pairs across diverse categories. Our dataset is constructed via a three-stage pipeline: (1) a general and input-aligned subject detection module, (2) large-scale cross-context subject retrieval from more than 53 million videos and 3 billion images, and (3) prior-guided identity verification to ensure visual consistency under contextual variation. Comprehensive experiments show that training with Phantom-Data significantly improves prompt alignment and visual quality while preserving identity consistency on par with in-pair baselines. |
| title | Phantom-Data : Towards a General Subject-Consistent Video Generation Dataset |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2506.18851 |