Guided Trajectory Optimization with Sparse Scaling for Test-Time Diffusion

2026-05-20

AI Summary

Existing test-time scaling methods rely on static noise pools, which hinder flexible exploration of denoising trajectories and limit the generative performance of diffusion models. To address this limitation, this work proposes RTS, a novel approach that integrates reward-guided noise optimization with a sparse scaling mechanism informed by PCA-based curvature analysis. By dynamically identifying and focusing on critical denoising steps during inference, RTS efficiently compresses the search space and refines the generation trajectory. Experimental results demonstrate that RTS achieves state-of-the-art performance, surpassing baseline methods by 15.6% on the GenEval Score and improving ImageReward by 60.4%.

Abstract

The efficient Test-Time Scaling (TTS) paradigm offers a promising perspective for enhancing the generation performance of diffusion models. However, current solutions are limited to a static, pre-defined noise pool and suffer from inflexible noise exploration across the denoising trajectory. To bridge this gap, we propose RTS, a novel Reward-guided Trajectory Scaling method to fully unlock the generative potential of diffusion models. Unlike existing methods, RTS facilitates the synthesis of refined, high-fidelity images via two core innovations: 1) a reward-guided noise optimization strategy to actively direct the search towards promising regions; and 2) a sparse test-time scaling framework together with a PCA-driven curvature analysis scheme to prioritize key intermediate steps in the entire denoising space, effectively compressing the search space. Experiments show our approach outperforms baselines by 15.6% across GenEval Score, and a 60.4% enhancement in ImageReward score, setting a new SOTA while providing a practical guideline for more effective test-time scaling across diffusion-specific architectures.

Problem

• Test-Time Scaling
• diffusion models
• noise exploration
• denoising trajectory
• generation performance

Innovation

• Reward-guided Optimization
• Sparse Test-Time Scaling
• Diffusion Models
• Trajectory Optimization
• PCA-driven Curvature Analysis