Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2512.16910 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912774414663680 |
|---|---|
| author | Rao, Qihang Zhang, Borui Zheng, Wenzhao Zhou, Jie Lu, Jiwen |
| author_facet | Rao, Qihang Zhang, Borui Zheng, Wenzhao Zhou, Jie Lu, Jiwen |
| contents | Recent advances in multimodal models highlight the pivotal role of image tokenization in high-resolution image generation. By compressing images into compact latent representations, tokenizers enable generative models to operate in lower-dimensional spaces, thereby improving computational efficiency and reducing complexity. Discrete tokenizers naturally align with the autoregressive paradigm but still lag behind continuous ones, limiting their adoption in multimodal systems. To address this, we propose \textbf{SFTok}, a discrete tokenizer that incorporates a multi-step iterative mechanism for precise reconstruction. By integrating \textbf{self-forcing guided visual reconstruction} and \textbf{debias-and-fitting training strategy}, SFTok resolves the training-inference inconsistency in multi-step process, significantly enhancing image reconstruction quality. At a high compression rate of only 64 tokens per image, SFTok achieves state-of-the-art reconstruction quality on ImageNet (rFID = 1.21) and demonstrates exceptional performance in class-to-image generation tasks (gFID = 2.29). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_16910 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | SFTok: Bridging the Performance Gap in Discrete Tokenizers Rao, Qihang Zhang, Borui Zheng, Wenzhao Zhou, Jie Lu, Jiwen Computer Vision and Pattern Recognition Machine Learning Recent advances in multimodal models highlight the pivotal role of image tokenization in high-resolution image generation. By compressing images into compact latent representations, tokenizers enable generative models to operate in lower-dimensional spaces, thereby improving computational efficiency and reducing complexity. Discrete tokenizers naturally align with the autoregressive paradigm but still lag behind continuous ones, limiting their adoption in multimodal systems. To address this, we propose \textbf{SFTok}, a discrete tokenizer that incorporates a multi-step iterative mechanism for precise reconstruction. By integrating \textbf{self-forcing guided visual reconstruction} and \textbf{debias-and-fitting training strategy}, SFTok resolves the training-inference inconsistency in multi-step process, significantly enhancing image reconstruction quality. At a high compression rate of only 64 tokens per image, SFTok achieves state-of-the-art reconstruction quality on ImageNet (rFID = 1.21) and demonstrates exceptional performance in class-to-image generation tasks (gFID = 2.29). |
| title | SFTok: Bridging the Performance Gap in Discrete Tokenizers |
| topic | Computer Vision and Pattern Recognition Machine Learning |
| url | https://arxiv.org/abs/2512.16910 |