The Year in AI Papers
The year/Independent research

Paper 2510.23588

FARMER: Flow AutoRegressive Transformer over Pixels

Guangting Zheng, Qinyu Zhao, Tao Yang, Fei Xiao, Zhijie Lin, Jie Wu, Jiajun Deng, Yanyong Zhang, Rui Zhu

Published
Oct 2025
Research lab
Independent
Citations
13
GitHub
Not linked

01 In brief

Summary

FARMER is a novel end-to-end generative framework that unifies Normalizing Flows (NF) and Autoregressive (AR) models for tractable likelihood estimation and high-quality image synthesis directly from raw pixels.

It employs an invertible autoregressive flow to transform images into latent sequences, whose distribution is modeled implicitly by an AR model.

To address redundancy in pixel-level modeling, FARMER introduces a self-supervised dimension reduction scheme that partitions NF latent channels into informative and redundant groups, enabling more effective and efficient AR modeling.

It also proposes a one-step distillation scheme to accelerate inference speed and a resampling-based classifier-free guidance algorithm to boost generation quality.

Experiments on ImageNet 256x256 show FARMER achieves competitive performance compared to existing pixel-based generative models while providing exact likelihoods and scalable training.

FARMER-1.9B achieves FID 3.60, outperforming JetFormer (FID 6.64) and STARFlow (FID 4.69).

The one-step distillation accelerates the NF reverse process by 22x and overall inference by 4x, while maintaining comparable quality.

02 From the paper

Abstract

Directly modeling the explicit likelihood of the raw data distribution is key topic in the machine learning area, which achieves the scaling successes in Large Language Models by autoregressive modeling. However, continuous AR modeling over visual pixel data suffer from extremely long sequences and high-dimensional spaces. In this paper, we present FARMER, a novel end-to-end generative framework that unifies Normalizing Flows (NF) and Autoregressive (AR) models for tractable likelihood estimation and high-quality image synthesis directly from raw pixels. FARMER employs an invertible autoregressive flow to transform images into latent sequences, whose distribution is modeled implicitly by an autoregressive model. To address the redundancy and complexity in pixel-level modeling, we propose a self-supervised dimension reduction scheme that partitions NF latent channels into informative and redundant groups, enabling more effective and efficient AR modeling. Furthermore, we design a one-step distillation scheme to significantly accelerate inference speed and introduce a resampling-based classifier-free guidance algorithm to boost image generation quality. Extensive experiments demonstrate that FARMER achieves competitive performance compared to existing pixel-based generative models while providing exact likelihoods and scalable training.