Image Tokenizer Needs Post-Training
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918141592862720 |
|---|---|
| author | Qiu, Kai Li, Xiang Chen, Hao Kuen, Jason Xu, Xiaohao Gu, Jiuxiang Luo, Yinyi Raj, Bhiksha Lin, Zhe Savvides, Marios |
| author_facet | Qiu, Kai Li, Xiang Chen, Hao Kuen, Jason Xu, Xiaohao Gu, Jiuxiang Luo, Yinyi Raj, Bhiksha Lin, Zhe Savvides, Marios |
| contents | Recent image generative models typically capture the image distribution in a pre-constructed latent space, relying on a frozen image tokenizer. However, there exists a significant discrepancy between the reconstruction and generation distribution, where current tokenizers only prioritize the reconstruction task that happens before generative training without considering the generation errors during sampling. In this paper, we comprehensively analyze the reason for this discrepancy in a discrete latent space, and, from which, we propose a novel tokenizer training scheme including both main-training and post-training, focusing on improving latent space construction and decoding respectively. During the main training, a latent perturbation strategy is proposed to simulate sampling noises, \ie, the unexpected tokens generated in generative inference. Specifically, we propose a plug-and-play tokenizer training scheme, which significantly enhances the robustness of tokenizer, thus boosting the generation quality and convergence speed, and a novel tokenizer evaluation metric, \ie, pFID, which successfully correlates the tokenizer performance to generation quality. During post-training, we further optimize the tokenizer decoder regarding a well-trained generative model to mitigate the distribution difference between generated and reconstructed tokens. With a $\sim$400M generator, a discrete tokenizer trained with our proposed main training achieves a notable 1.60 gFID and further obtains 1.36 gFID with the additional post-training. Further experiments are conducted to broadly validate the effectiveness of our post-training strategy on off-the-shelf discrete and continuous tokenizers, coupled with autoregressive and diffusion-based generators. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_12474 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Image Tokenizer Needs Post-Training Qiu, Kai Li, Xiang Chen, Hao Kuen, Jason Xu, Xiaohao Gu, Jiuxiang Luo, Yinyi Raj, Bhiksha Lin, Zhe Savvides, Marios Computer Vision and Pattern Recognition Recent image generative models typically capture the image distribution in a pre-constructed latent space, relying on a frozen image tokenizer. However, there exists a significant discrepancy between the reconstruction and generation distribution, where current tokenizers only prioritize the reconstruction task that happens before generative training without considering the generation errors during sampling. In this paper, we comprehensively analyze the reason for this discrepancy in a discrete latent space, and, from which, we propose a novel tokenizer training scheme including both main-training and post-training, focusing on improving latent space construction and decoding respectively. During the main training, a latent perturbation strategy is proposed to simulate sampling noises, \ie, the unexpected tokens generated in generative inference. Specifically, we propose a plug-and-play tokenizer training scheme, which significantly enhances the robustness of tokenizer, thus boosting the generation quality and convergence speed, and a novel tokenizer evaluation metric, \ie, pFID, which successfully correlates the tokenizer performance to generation quality. During post-training, we further optimize the tokenizer decoder regarding a well-trained generative model to mitigate the distribution difference between generated and reconstructed tokens. With a $\sim$400M generator, a discrete tokenizer trained with our proposed main training achieves a notable 1.60 gFID and further obtains 1.36 gFID with the additional post-training. Further experiments are conducted to broadly validate the effectiveness of our post-training strategy on off-the-shelf discrete and continuous tokenizers, coupled with autoregressive and diffusion-based generators. |
| title | Image Tokenizer Needs Post-Training |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2509.12474 |