GANonymizer: Image Anonymization Method Integrating Object Detection and Generative Adversarial Network

Tanimura, Tomoki; Kawano, Mahiru; Yonezawa, Takeshi; Nakazawa, Jin

doi:10.1007/978-3-030-28925-6_10

Cited by 1 publication

(1 citation statement)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…1) Our Contributions: Our proposed method follows a two step approach by first identifying and masking out faces in individual frames and then inpainting the missing content with an artificially generated new face using a temporally coherent Generative Adversarial Network [8]. We demonstrate that using such a method not only allows to generate consistent faces across several image frames within a video sequence, but also leads to more natural looking faces and a higher FID score [10] even in single 1 Download script is available at: https://github.com/cvims/ jagan images compared to state of the art methods [34] [13] [39]. To ensure a full anonymization without leaking any information from the original face, our approach avoids using any facial characteristics or features such as landmarks.…”

Section: Introductionmentioning

confidence: 99%

Temporally coherent video anonymization through GAN inpainting

Balaji,

Blies,

Göri

et al. 2021

Preprint

View full text Add to dashboard Cite

This work tackles the problem of temporally coherent face anonymization in natural video streams. We propose JaGAN, a two-stage system starting with detecting and masking out faces with black image patches in all individual frames of the video. The second stage leverages a privacy-preserving Video Generative Adversarial Network designed to inpaint the missing image patches with artificially generated faces. Our initial experiments reveal that image based generative models are not capable of inpainting patches showing temporal coherent appearance across neighboring video frames. To address this issue we introduce a newly curated video collection, which is made publicly available for the research community along with this paper 1 . We also introduce the Identity Invariance Score IdI as a means to quantify temporal coherency between neighboring frames.

show abstract