Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.19488 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911410617843712 |
|---|---|
| author | Han, Yizhao Shi, Tianxing Wang, Zhao Xu, Zifan Pu, Zhiyuan Li, Mingxiao Zhang, Qian Yin, Wei Long, Xiao-Xiao |
| author_facet | Han, Yizhao Shi, Tianxing Wang, Zhao Xu, Zifan Pu, Zhiyuan Li, Mingxiao Zhang, Qian Yin, Wei Long, Xiao-Xiao |
| contents | Autoregressive (AR) architectures have achieved significant successes in LLMs, inspiring explorations for video generation. In LLMs, top-p/top-k sampling strategies work exceptionally well: language tokens have high semantic density and low redundancy, so a fixed size of token candidates already strikes a balance between semantic accuracy and generation diversity. In contrast, video tokens have low semantic density and high spatio-temporal redundancy. This mismatch makes static top-k/top-p strategies ineffective for video decoders: they either introduce unnecessary randomness for low-uncertainty regions (static backgrounds) or get stuck in early errors for high-uncertainty regions (foreground objects). Prediction errors will accumulate as more frames are generated and eventually severely degrade long-horizon quality. To address this, we propose Entropy-Guided k-Guard (ENkG) sampling, a simple yet effective strategy that adapts sampling to token-wise dispersion, quantified by the entropy of each token's predicted distribution. ENkG uses adaptive token candidate sizes: for low-entropy regions, it employs fewer candidates to suppress redundant noise and preserve structural integrity; for high-entropy regions, it uses more candidates to mitigate error compounding. ENkG is model-agnostic, training-free, and adds negligible overhead. Experiments demonstrate consistent improvements in perceptual quality and structural stability compared to static top-k/top-p strategies. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_19488 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Entropy-Guided k-Guard Sampling for Long-Horizon Autoregressive Video Generation Han, Yizhao Shi, Tianxing Wang, Zhao Xu, Zifan Pu, Zhiyuan Li, Mingxiao Zhang, Qian Yin, Wei Long, Xiao-Xiao Computer Vision and Pattern Recognition Autoregressive (AR) architectures have achieved significant successes in LLMs, inspiring explorations for video generation. In LLMs, top-p/top-k sampling strategies work exceptionally well: language tokens have high semantic density and low redundancy, so a fixed size of token candidates already strikes a balance between semantic accuracy and generation diversity. In contrast, video tokens have low semantic density and high spatio-temporal redundancy. This mismatch makes static top-k/top-p strategies ineffective for video decoders: they either introduce unnecessary randomness for low-uncertainty regions (static backgrounds) or get stuck in early errors for high-uncertainty regions (foreground objects). Prediction errors will accumulate as more frames are generated and eventually severely degrade long-horizon quality. To address this, we propose Entropy-Guided k-Guard (ENkG) sampling, a simple yet effective strategy that adapts sampling to token-wise dispersion, quantified by the entropy of each token's predicted distribution. ENkG uses adaptive token candidate sizes: for low-entropy regions, it employs fewer candidates to suppress redundant noise and preserve structural integrity; for high-entropy regions, it uses more candidates to mitigate error compounding. ENkG is model-agnostic, training-free, and adds negligible overhead. Experiments demonstrate consistent improvements in perceptual quality and structural stability compared to static top-k/top-p strategies. |
| title | Entropy-Guided k-Guard Sampling for Long-Horizon Autoregressive Video Generation |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.19488 |