Rendering-Refined Stable Diffusion for Privacy Compliant Synthetic Data

Kartik Patwari, David Schneider, Xiaoxiao Sun, Chen-Nee Chuah, Lingjuan Lyu, Vivek Sharma · 2024

Read Rendering-Refined Stable Diffusion for Privacy Compliant Synthetic Data by Kartik Patwari, David Schneider, Xiaoxiao Sun, Chen-Nee Chuah, Lingjuan Lyu, Vivek Sharma (2024) 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-2412-06248.txt.

Description

Six researchers present a pipeline to anonymize people in image data sets in a 2024 paper. Privacy laws demand identity protection, yet crude tools like blur wreck image value, most of all for human scenes. Their method joins 3D rendering with a diffusion image model.

Rendering locks the body's posture while the diffusion step repaints identity traits from a text prompt. That split beats plain diffusion, which loses posture, and older GAN tools, which lack realism and control. The team also built a framework to grade privacy and utility by human judgment. Models trained on their fake data beat real data training on detection tasks in their tests.

Catalog record

CREATORKartik Patwari, David Schneider, Xiaoxiao Sun, Chen-Nee Chuah, Lingjuan Lyu, Vivek Sharma
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-06248

Browse the full library · About the Tower of Babel