2025· Neural Information Processing Systems· pp. 114865-114890· 6 citations· ⚡ 1 influential· 64 references
Computer Science
TL;DR
This work proposes FreqExit, an integrated loss design that enables dynamic inference in VAR without altering its architecture or compromising output quality, and designs targeted mechanisms that guide the model to learn more effectively through frequency-aware supervision.
Abstract
. Abstract Visual AutoRegressive (VAR) modeling employs a next-scale decoding paradigm that progresses from coarse structures to fine details. While enhancing fidelity and scalability, this approach challenges two fundamental assumptions of conventional dynamic inference: semantic stability (intermediate outputs approximating final results) and monotonic locality (smooth representation evolution across layers), which renders existing dynamic inference methods ineffective for VAR models. To address this challenge, we propose FreqExit , an integrated loss design that enables dynamic inference in VAR without altering its architecture or compromising output quality. FreqExit is based on a key insight: high-frequency details are crucial for perceptual quality and tend to emerge only in later decoding stages. Leveraging this insight, we design targeted mechanisms that guide the model to learn more effectively through frequency-aware supervision. The proposed framework consists of three components: (1) a curriculum-based supervision strategy with progressive layer dropout and early exit loss; (2) a wavelet-domain high-frequency consistency loss that aligns spectral content across different generation steps; and (3) a lightweight self-supervised frequency-gated
Visual Autoregressive Models (VAR) generate images through next-scale prediction, producing all tokens within each scale in parallel. We show that this parallel decoding constitutes a mean-field-style approximation that discards spatial dependencies among same-scale tokens, causing locally incoherent samples regardless of backbone capacity -- a limitation of the decoding rule. Addressing this limitation, we introduce the Logit Refiner, a lightweight autoregressive module that restores intra-scale dependencies by sequentially sampling tokens conditioned on frozen backbone features. Adding only ~10% parameters and less than 5% of the base model's training compute, it plugs into any pretrained VAR checkpoint without retraining. Controlled ablations isolate joint intra-scale sampling -- rather than additional capacity or training -- as the critical ingredient. Across backbones from 310M to 2B parameters on class-conditional ImageNet 256x256, the refiner consistently improves generation quality, enabling a 1.1B-parameter model to surpass one twice its size. The approach further generalizes to text-to-image generation, confirming that the mean-field bottleneck persists across VAR variants and is effectively alleviated by our method. Project page: https://compvis.github.io/logit-refiner/
Meimingwei Li, Stefan Andreas Baumann, Felix Krause et al.· 0 citations
This work proposes a Scale-wise Conditional Decoding mechanism, which projects semantic signals from a frozen vision encoder into scale-specific layers of the VAR backbone, and introduces a Unified Control Alignment strategy (SCALAR-Uni) to handle diverse control modalities within a single projection space.
Ryan Xu, Dongyang Jin, Shawn Chen et al.· International Journal of Com...· 0 citations
Autoregressive (AR) image generation has shown strong potential for scalable high-fidelity synthesis by modeling images as discrete token sequences. However, traditional next token prediction (NTP) continues to suffer from sparse and myopic supervision, insufficiently discriminative representations, and high training cost caused by dense computation over the full token sequence. To address these issues, we propose multi-token autoregressive (MTAR), a unified training framework that improves autoregressive image generation from three aspects: prediction objectives, representation regularization, and training efficiency. Specifically, MTAR introduces multi-token prediction (MTP) to alleviate the sparsity and myopia of traditional NTP by imposing joint supervision on multiple future tokens; employs token-level contrastive regularization (TCR) to explicitly enhance the separability of sampled token representations and thereby improve representation discriminability; and incorporates semantic dropping (SD) as a semantics-aware training acceleration strategy to reduce redundant computation on low-information tokens while preserving informative learning signals. All three components are applied only during training and introduce no additional overhead during autoregressive inference. On ImageNet, MTAR achieves a better balance between generation quality and training efficiency. Compared with LlamaGen, MTAR achieves up to 0.95 lower FID and 39\% faster training. Moreover, even with only 1/3 of the training iterations, it still attains performance comparable to or better than the baseline, substantially reducing training time.
Guo Niu, Xiongfei Yao, Teng Wang et al.· 0 citations
This paper explores post-training reinforcement learning (RL), specifically GRPO, to directly align autoregressive perception models with their evaluation metrics, and designs an RL framework that addresses perception-specific challenges: reward design for set-structured outputs and multi-head sampling control.
Sofian Chaybouti, Yasser Dahou, N. Huynh et al.· 0 citations
The new version of AlayaWorld substantially revise how conditioning signals are represented and integrated into the model, replacing the previous depth-warping-based spatial memory with a streaming 3D point-cache renderer.
AlayaWorld Team Kaipeng Zhang, Chuanhao Li, Y. Zhan et al.· 1 citation· ⚡1
This work establishes the first connection between autoregressive video drifting and model internal representations, adopts erank as a quantitative metric for error accumulation, reveals counterintuitive scaling limitations for video world models, and presents a simple yet effective regularization strategy to improve long video generation robustness.
