English

SNCE: Geometry-Aware Supervision for Scalable Discrete Image Generation

Computer Vision and Pattern Recognition 2026-03-17 v1

Abstract

Recent advancements in discrete image generation showed that scaling the VQ codebook size significantly improves reconstruction fidelity. However, training generative models with a large VQ codebook remains challenging, typically requiring larger model size and a longer training schedule. In this work, we propose Stochastic Neighbor Cross Entropy Minimization (SNCE), a novel training objective designed to address the optimization challenges of large-codebook discrete image generators. Instead of supervising the model with a hard one-hot target, SNCE constructs a soft categorical distribution over a set of neighboring tokens. The probability assigned to each token is proportional to the proximity between its code embedding and the ground-truth image embedding, encouraging the model to capture semantically meaningful geometric structure in the quantized embedding space. We conduct extensive experiments across class-conditional ImageNet-256 generation, large-scale text-to-image synthesis, and image editing tasks. Results show that SNCE significantly improves convergence speed and overall generation quality compared to standard cross-entropy objectives.

Keywords

Cite

@article{arxiv.2603.15150,
  title  = {SNCE: Geometry-Aware Supervision for Scalable Discrete Image Generation},
  author = {Shufan Li and Jiuxiang Gu and Kangning Liu and Zhe Lin and Aditya Grover and Jason Kuen},
  journal= {arXiv preprint arXiv:2603.15150},
  year   = {2026}
}

Comments

21 pages, 4 figures

R2 v1 2026-07-01T11:22:06.447Z