Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2403.06951 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914710732931072 |
|---|---|
| author | Qi, Tianhao Fang, Shancheng Wu, Yanze Xie, Hongtao Liu, Jiawei Chen, Lang He, Qian Zhang, Yongdong |
| author_facet | Qi, Tianhao Fang, Shancheng Wu, Yanze Xie, Hongtao Liu, Jiawei Chen, Lang He, Qian Zhang, Yongdong |
| contents | The diffusion-based text-to-image model harbors immense potential in transferring reference style. However, current encoder-based approaches significantly impair the text controllability of text-to-image models while transferring styles. In this paper, we introduce DEADiff to address this issue using the following two strategies: 1) a mechanism to decouple the style and semantics of reference images. The decoupled feature representations are first extracted by Q-Formers which are instructed by different text descriptions. Then they are injected into mutually exclusive subsets of cross-attention layers for better disentanglement. 2) A non-reconstructive learning method. The Q-Formers are trained using paired images rather than the identical target, in which the reference image and the ground-truth image are with the same style or semantics. We show that DEADiff attains the best visual stylization results and optimal balance between the text controllability inherent in the text-to-image model and style similarity to the reference image, as demonstrated both quantitatively and qualitatively. Our project page is https://tianhao-qi.github.io/DEADiff/. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_06951 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | DEADiff: An Efficient Stylization Diffusion Model with Disentangled Representations Qi, Tianhao Fang, Shancheng Wu, Yanze Xie, Hongtao Liu, Jiawei Chen, Lang He, Qian Zhang, Yongdong Computer Vision and Pattern Recognition The diffusion-based text-to-image model harbors immense potential in transferring reference style. However, current encoder-based approaches significantly impair the text controllability of text-to-image models while transferring styles. In this paper, we introduce DEADiff to address this issue using the following two strategies: 1) a mechanism to decouple the style and semantics of reference images. The decoupled feature representations are first extracted by Q-Formers which are instructed by different text descriptions. Then they are injected into mutually exclusive subsets of cross-attention layers for better disentanglement. 2) A non-reconstructive learning method. The Q-Formers are trained using paired images rather than the identical target, in which the reference image and the ground-truth image are with the same style or semantics. We show that DEADiff attains the best visual stylization results and optimal balance between the text controllability inherent in the text-to-image model and style similarity to the reference image, as demonstrated both quantitatively and qualitatively. Our project page is https://tianhao-qi.github.io/DEADiff/. |
| title | DEADiff: An Efficient Stylization Diffusion Model with Disentangled Representations |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2403.06951 |