Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy
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_ | 1866908601348521984 |
|---|---|
| author | Ma, Xiaoxiao Zhao, Feng Ling, Pengyang Qiu, Haibo Wei, Zhixiang Yu, Hu Huang, Jie Zeng, Zhixiong Ma, Lin |
| author_facet | Ma, Xiaoxiao Zhao, Feng Ling, Pengyang Qiu, Haibo Wei, Zhixiang Yu, Hu Huang, Jie Zeng, Zhixiong Ma, Lin |
| contents | In this work, we first revisit the sampling issues in current autoregressive (AR) image generation models and identify that image tokens, unlike text tokens, exhibit lower information density and non-uniform spatial distribution. Accordingly, we present an entropy-informed decoding strategy that facilitates higher autoregressive generation quality with faster synthesis speed. Specifically, the proposed method introduces two main innovations: 1) dynamic temperature control guided by spatial entropy of token distributions, enhancing the balance between content diversity, alignment accuracy, and structural coherence in both mask-based and scale-wise models, without extra computational overhead, and 2) entropy-aware acceptance rules in speculative decoding, achieving near-lossless generation at about 85\% of the inference cost of conventional acceleration methods. Extensive experiments across multiple benchmarks using diverse AR image generation models demonstrate the effectiveness and generalizability of our approach in enhancing both generation quality and sampling speed. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_09012 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy Ma, Xiaoxiao Zhao, Feng Ling, Pengyang Qiu, Haibo Wei, Zhixiang Yu, Hu Huang, Jie Zeng, Zhixiong Ma, Lin Computer Vision and Pattern Recognition In this work, we first revisit the sampling issues in current autoregressive (AR) image generation models and identify that image tokens, unlike text tokens, exhibit lower information density and non-uniform spatial distribution. Accordingly, we present an entropy-informed decoding strategy that facilitates higher autoregressive generation quality with faster synthesis speed. Specifically, the proposed method introduces two main innovations: 1) dynamic temperature control guided by spatial entropy of token distributions, enhancing the balance between content diversity, alignment accuracy, and structural coherence in both mask-based and scale-wise models, without extra computational overhead, and 2) entropy-aware acceptance rules in speculative decoding, achieving near-lossless generation at about 85\% of the inference cost of conventional acceleration methods. Extensive experiments across multiple benchmarks using diverse AR image generation models demonstrate the effectiveness and generalizability of our approach in enhancing both generation quality and sampling speed. |
| title | Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2510.09012 |