Learning Visual Generative Priors without Text
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910889254322176 |
|---|---|
| author | Ma, Shuailei Zheng, Kecheng Wei, Ying Wu, Wei Lu, Fan Zhang, Yifei Xie, Chen-Wei Gong, Biao Zhu, Jiapeng Shen, Yujun |
| author_facet | Ma, Shuailei Zheng, Kecheng Wei, Ying Wu, Wei Lu, Fan Zhang, Yifei Xie, Chen-Wei Gong, Biao Zhu, Jiapeng Shen, Yujun |
| contents | Although text-to-image (T2I) models have recently thrived as visual generative priors, their reliance on high-quality text-image pairs makes scaling up expensive. We argue that grasping the cross-modality alignment is not a necessity for a sound visual generative prior, whose focus should be on texture modeling. Such a philosophy inspires us to study image-to-image (I2I) generation, where models can learn from in-the-wild images in a self-supervised manner. We first develop a pure vision-based training framework, Lumos, and confirm the feasibility and the scalability of learning I2I models. We then find that, as an upstream task of T2I, our I2I model serves as a more foundational visual prior and achieves on-par or better performance than existing T2I models using only 1/10 text-image pairs for fine-tuning. We further demonstrate the superiority of I2I priors over T2I priors on some text-irrelevant visual generative tasks, like image-to-3D and image-to-video. Our project page is available at https://ant-research.github.io/lumos. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_07767 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Learning Visual Generative Priors without Text Ma, Shuailei Zheng, Kecheng Wei, Ying Wu, Wei Lu, Fan Zhang, Yifei Xie, Chen-Wei Gong, Biao Zhu, Jiapeng Shen, Yujun Computer Vision and Pattern Recognition Although text-to-image (T2I) models have recently thrived as visual generative priors, their reliance on high-quality text-image pairs makes scaling up expensive. We argue that grasping the cross-modality alignment is not a necessity for a sound visual generative prior, whose focus should be on texture modeling. Such a philosophy inspires us to study image-to-image (I2I) generation, where models can learn from in-the-wild images in a self-supervised manner. We first develop a pure vision-based training framework, Lumos, and confirm the feasibility and the scalability of learning I2I models. We then find that, as an upstream task of T2I, our I2I model serves as a more foundational visual prior and achieves on-par or better performance than existing T2I models using only 1/10 text-image pairs for fine-tuning. We further demonstrate the superiority of I2I priors over T2I priors on some text-irrelevant visual generative tasks, like image-to-3D and image-to-video. Our project page is available at https://ant-research.github.io/lumos. |
| title | Learning Visual Generative Priors without Text |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2412.07767 |