Abstract
Some personally identifiable information (PII) redaction methods for generative image applications can have high latency and computational costs, which may limit their use in interactive contexts. A dual-pathway architecture may be used to process image modification requests. A synchronous, low-latency pathway can detect and obfuscate potential PII before a generative model concurrently inpaints the obfuscated areas and applies user modifications. Simultaneously, an asynchronous, offline pathway can generate a more thoroughly redacted version of the source image and store it for future use. This hybrid approach can provide a low-latency response suitable for on-demand image generation, while progressively creating higher-quality redacted assets for frequently requested content, thus balancing processing speed with the quality of the output.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Leung, Mira, "Dual-Pathway System for Low-Latency PII Redaction in Generative Image Modification", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10820