Ctrl-Z Sampling: Scaling Diffusion Sampling with Controlled Random Zigzag Explorations

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mao, Shunqi, Guo, Wei, Zhang, Chaoyi, Long, Jieting, Xie, Ke, Cai, Weidong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910064048078848
author Mao, Shunqi
Guo, Wei
Zhang, Chaoyi
Long, Jieting
Xie, Ke
Cai, Weidong
author_facet Mao, Shunqi
Guo, Wei
Zhang, Chaoyi
Long, Jieting
Xie, Ke
Cai, Weidong
contents Diffusion models generate conditional samples by progressively denoising Gaussian noise, yet the denoising trajectory can stall at visually plausible but low-quality outcomes with conditional misalignment or structural artifacts. We interpret this behavior as local optima in a surrogate quality landscape: Once early denoising commits to a suboptimal global structure, later steps mainly sharpen details and seldom correct the underlying mistake. While existing inference-time approaches explore alternative diffusion states via re-noising with fixed strength or direction, they exhibit limited capacity to escape steep quality plateaus. We propose Controlled Random Zigzag Sampling (Ctrl-Z Sampling),a scalable sampling strategy that detects plateaus in quality landscape via a surrogate score, and allocates exploration only when a plateau is detected. Upon detection, Ctrl-Z Sampling rolls back to noisier states, samples a set of alternative continuations, and updates the trajectory when a candidate improves the score, otherwise escalating the exploration depth to escape the current plateau. The proposed method is model-agnostic and broadly compatible with existing diffusion frameworks. Experiments show that Ctrl-Z Sampling consistently improves generation quality over other inference-time scaling samplers across different NFE budgets, offering a scalable compute-quality trade-off.
format Preprint
id arxiv_https___arxiv_org_abs_2506_20294
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Ctrl-Z Sampling: Scaling Diffusion Sampling with Controlled Random Zigzag Explorations
Mao, Shunqi
Guo, Wei
Zhang, Chaoyi
Long, Jieting
Xie, Ke
Cai, Weidong
Computer Vision and Pattern Recognition
Diffusion models generate conditional samples by progressively denoising Gaussian noise, yet the denoising trajectory can stall at visually plausible but low-quality outcomes with conditional misalignment or structural artifacts. We interpret this behavior as local optima in a surrogate quality landscape: Once early denoising commits to a suboptimal global structure, later steps mainly sharpen details and seldom correct the underlying mistake. While existing inference-time approaches explore alternative diffusion states via re-noising with fixed strength or direction, they exhibit limited capacity to escape steep quality plateaus. We propose Controlled Random Zigzag Sampling (Ctrl-Z Sampling),a scalable sampling strategy that detects plateaus in quality landscape via a surrogate score, and allocates exploration only when a plateau is detected. Upon detection, Ctrl-Z Sampling rolls back to noisier states, samples a set of alternative continuations, and updates the trajectory when a candidate improves the score, otherwise escalating the exploration depth to escape the current plateau. The proposed method is model-agnostic and broadly compatible with existing diffusion frameworks. Experiments show that Ctrl-Z Sampling consistently improves generation quality over other inference-time scaling samplers across different NFE budgets, offering a scalable compute-quality trade-off.
title Ctrl-Z Sampling: Scaling Diffusion Sampling with Controlled Random Zigzag Explorations
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.20294