RODS: Robust Optimization Inspired Diffusion Sampling for Detecting and Reducing Hallucination in Generative Models
Yiqi Tian *, Pengfei Jin *, Mingze Yuan, Na Li, Bo Zeng †, Quanzheng Li † · 2025
Read RODS: Robust Optimization Inspired Diffusion Sampling for Detecting and Reducing Hallucination in Generative Models by Yiqi Tian *, Pengfei Jin *, Mingze Yuan, Na Li, Bo Zeng †, Quanzheng Li † (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-12201.txt.
Description
Tian and five co-authors present RODS, a sampler that detects and corrects hallucinated steps in diffusion image generation, in a paper from twenty twenty-five. Diffusion models lead generative modeling, yet score errors inject false details during sampling. The team reframes sampling as optimization over a loss landscape. Geometric cues then flag risky steps before artifacts form.
RODS smooths trajectories and adapts perturbations at flagged steps. No retraining is needed, and extra inference cost stays minimal. Tests on faces, animals, and hands data sets preserve quality and diversity. Over seventy percent of hallucinated samples are detected. More than a quarter are fully corrected without new artifacts. Code is released openly.
Catalog record
| CREATOR | Yiqi Tian *, Pengfei Jin *, Mingze Yuan, Na Li, Bo Zeng †, Quanzheng Li † |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-12201