Next-Dense-Stride Prediction for Multimodal Autoregressive Visual Modeling

2026-07-10

AI Summary

This work addresses the inefficiencies of conventional autoregressive image generation—specifically, the slow inference imposed by raster-scan ordering and the high computational cost of multi-scale approaches—by introducing DenseAR, a novel framework. DenseAR operates on a single-scale latent grid and features a pioneering next-dense-stride prediction mechanism that enables progressively denser parallel token prediction, facilitating efficient autoregressive generation from global structure to fine details. By integrating a compact tokenizer, dense-stride traversal, and multi-task conditional modeling, DenseAR unifies natural image synthesis and multimodal medical imaging tasks within a single backbone architecture for the first time. Experiments demonstrate its superior performance on ImageNet over existing single-grid and multi-scale baselines in terms of FID and Inception Score, as well as successful cross-modal translation, conditional generation, and tumor segmentation on multi-contrast brain MRI.

Abstract

We introduce DenseAR, a new generative paradigm that reformulates autoregressive image generation as coarse-to-fine next-dense-stride prediction using a compact single-scale tokenizer. Our key insight is that traversing a single-scale latent grid with progressively denser strides naturally captures the transition from global structure to fine detail. This addresses two limitations of existing autoregressive models at once: the slow inference of raster-order autoregression, which DenseAR avoids by predicting multiple tokens in parallel, and the heavy cost of multi-scale approaches, which need long, multi-resolution token sequences to achieve coarse-to-fine prediction. Building on our efficient framework and the flexibility of autoregressive modeling, we further extend DenseAR to a unified model that handles multiple modalities and imaging tasks within a single backbone. We validate DenseAR on both medical and natural images. On multi-contrast brain MRI, a single DenseAR model unifies cross-modal translation, modality-conditioned generation, and tumor segmentation, while remaining competitive with task-specific methods. On ImageNet, DenseAR improves class-conditional generation quality (FID and IS) over both a single-grid baseline without stride ordering and a multi-scale tokenizer-based baseline.

Problem

• autoregressive modeling
• image generation
• coarse-to-fine prediction
• multi-scale approaches
• raster-order autoregression

Innovation

• next-dense-stride prediction
• single-scale tokenizer
• coarse-to-fine autoregression
• multimodal unified modeling
• parallel token generation