Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2603.25249 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917362904596480 |
|---|---|
| author | Li, Qingfeng Zhang, Haoxian He, Xu Tang, Songlin Fang, Zhixue Liu, Xiaoqiang Li, Pengfei Wan Guoqi |
| author_facet | Li, Qingfeng Zhang, Haoxian He, Xu Tang, Songlin Fang, Zhixue Liu, Xiaoqiang Li, Pengfei Wan Guoqi |
| contents | Visual tokenizers play a central role in latent image generation by bridging high-dimensional images and tractable generative modeling. However, most existing tokenizers are still trained with reconstruction-dominated objectives, which often yield latent representations that are only weakly grounded in high-level semantics. Recent approaches improve semantic alignment, but typically treat semantic signals as auxiliary regularization rather than making them functionally necessary for representation learning. We propose SMAP, a SeMantic-Aware Prefix tokenizer that injects class-level semantic conditions into a query-based 1D tokenization framework. To make semantics indispensable during training, SMAP introduces a tail token dropping strategy, which forces semantic conditions and early latent prefixes to bear increasing responsibility under progressively reduced token budgets. To verify that the resulting latent space is useful for generation rather than reconstruction alone, we further introduce CARD, a hybrid Causal AutoRegressive--Diffusion generator. Extensive experiments on ImageNet show that SMAP consistently improves reconstruction quality across discrete and continuous tokenization settings, and that its semantically grounded latent space yields strong downstream generation performance under compact token budgets. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_25249 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Semantic-Aware Prefix Learning for Token-Efficient Image Generation Li, Qingfeng Zhang, Haoxian He, Xu Tang, Songlin Fang, Zhixue Liu, Xiaoqiang Li, Pengfei Wan Guoqi Computer Vision and Pattern Recognition Visual tokenizers play a central role in latent image generation by bridging high-dimensional images and tractable generative modeling. However, most existing tokenizers are still trained with reconstruction-dominated objectives, which often yield latent representations that are only weakly grounded in high-level semantics. Recent approaches improve semantic alignment, but typically treat semantic signals as auxiliary regularization rather than making them functionally necessary for representation learning. We propose SMAP, a SeMantic-Aware Prefix tokenizer that injects class-level semantic conditions into a query-based 1D tokenization framework. To make semantics indispensable during training, SMAP introduces a tail token dropping strategy, which forces semantic conditions and early latent prefixes to bear increasing responsibility under progressively reduced token budgets. To verify that the resulting latent space is useful for generation rather than reconstruction alone, we further introduce CARD, a hybrid Causal AutoRegressive--Diffusion generator. Extensive experiments on ImageNet show that SMAP consistently improves reconstruction quality across discrete and continuous tokenization settings, and that its semantically grounded latent space yields strong downstream generation performance under compact token budgets. |
| title | Semantic-Aware Prefix Learning for Token-Efficient Image Generation |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.25249 |