DeepPrivacy2: Towards Realistic Full-Body Anonymization

Hukkelås, Håkon; Lindseth, Frank

doi:10.1109/wacv56688.2023.00138

Cited by 18 publications

(18 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…preserving facial expression) but offer no formal guarantee of removing the original identity from the image, making them vulnerable to adversarial attacks. A few methods explore anonymizing the full-body [4,23,26,38], where the current state-of-the-art [23,24] can generate convincing full-bodies given sparse keypoints [24] or dense pose annotations [23]. Finally, some methods insert adversarial perturbation in the image, which is invisible to the human eye but able to fool face recognition systems [46].…”

Section: Related Workmentioning

confidence: 99%

“…Full-Body Anonymization Since all benchmark datasets include annotated instance segmentations, we use these to define the full-body anonymization region. To compensate for annotations where the segmentations don't fully encompass the body (often segmentation does not include bordering pixels), we slightly dilate the segmentation following [23].…”

Section: Anonymization Regionmentioning

confidence: 99%

“…For realistic anonymization, we employ pre-trained models from DeepPrivacy2 [23]. Note that DeepPrivacy2 anonymizes by inpainting (illustrated in Fig.…”

Section: Realistic Anonymizationmentioning

confidence: 99%

“…gait, clothes, ear, body shape), often requiring fullbody anonymization to protect privacy. A few studies explore the impact of full-body anonymization [23,26], where they find it to improve over traditional methods. However, they rely on automatic detection methods, which opens the question if the performance degradation is due to detection errors or the anonymization model.…”

Section: Introductionmentioning

confidence: 99%

“…In this paper, we focus on key computer vision tasks related to autonomous vehicles, namely instance segmentation and human pose estimation. We evaluate the full-body and face anonymization models built in DeepPrivacy2 [23] and compare realistic anonymization to traditional methods. See https://github.com/hukkelas/deep_ privacy2 / blob / master / docs / anonymizing _ datasets.md to reproduce our experiments.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Image Inpainting with Learnable Feature Imputation

Hukkelås

Lindseth

Mester

2021

Lecture Notes in Computer Science

View full text Add to dashboard Cite

A regular convolution layer applying a filter in the same way over known and unknown areas causes visual artifacts in the inpainted image. Several studies address this issue with feature re-normalization on the output of the convolution. However, these models use a significant amount of learnable parameters for feature re-normalization [ 41 , 48 ], or assume a binary representation of the certainty of an output [ 11 , 26 ]. We propose (layer-wise) feature imputation of the missing input values to a convolution. In contrast to learned feature re-normalization [ 41 , 48 ], our method is efficient and introduces a minimal number of parameters. Furthermore, we propose a revised gradient penalty for image inpainting, and a novel GAN architecture trained exclusively on adversarial loss. Our quantitative evaluation on the FDF dataset reflects that our revised gradient penalty and alternative convolution improves generated image quality significantly. We present comparisons on CelebA-HQ and Places2 to current state-of-the-art to validate our model. (Code is available at: github.com/hukkelas/DeepPrivacy . Supplementary material can be downloaded from: folk.ntnu.no/haakohu/GCPR_supplementary.pdf ) Electronic supplementary material The online version of this chapter (10.1007/978-3-030-71278-5_28) contains supplementary material, which is available to authorized users.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Anonymization Regionmentioning

confidence: 99%

“…For realistic anonymization, we employ pre-trained models from DeepPrivacy2 [23]. Note that DeepPrivacy2 anonymizes by inpainting (illustrated in Fig.…”

Section: Realistic Anonymizationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Image Inpainting with Learnable Feature Imputation

Hukkelås

Lindseth

Mester

2021

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

Computational facial analysis for rare Mendelian disorders

Hsieh

Krawitz

2023

American J of Med Genetics Pt C

View full text Add to dashboard Cite

With the advances in computer vision, computational facial analysis has become a powerful and effective tool for diagnosing rare disorders. This technology, also called next‐generation phenotyping (NGP), has progressed significantly over the last decade. This review paper will introduce three key NGP approaches. In 2014, Ferry et al. first presented Clinical Face Phenotype Space (CFPS) trained on eight syndromes. After 5 years, Gurovich et al. proposed DeepGestalt, a deep convolutional neural network trained on more than 21,000 patient images with 216 disorders. It was considered a state‐of‐the‐art disorder classification framework. In 2022, Hsieh et al. developed GestaltMatcher to support the ultra‐rare and novel disorders not supported in DeepGestalt. It further enabled the analysis of facial similarity presented in a given cohort or multiple disorders. Moreover, this article will present the usage of NGP for variant prioritization and facial gestalt delineation. Although NGP approaches have proven their capability in assisting the diagnosis of many disorders, many limitations remain. This article will introduce two future directions to address two main limitations: enabling the global collaboration for a medical imaging database that fulfills the FAIR principles and synthesizing patient images to protect patient privacy. In the end, with more and more NGP approaches emerging, we envision that the NGP technology can assist clinicians and researchers in diagnosing patients and analyzing disorders in multiple directions in the near future.

show abstract

Privacy-preserving visual analysis: training video obfuscation models without sensitive labels

De Coninck,

Wang,

Leroux

et al. 2024

Appl Intell

View full text Add to dashboard Cite

Visual analysis tasks, including crowd management, often require resource-intensive machine learning models, posing challenges for deployment on edge hardware. Consequently, cloud computing emerges as a prevalent solution. To address privacy concerns associated with offloading video data to remote cloud platforms, we present a novel approach using adversarial training to develop a lightweight obfuscator neural network. Our method focuses on pedestrian detection as an example of visual analysis, allowing the transformation of video frames on the camera itself to retain only essential information for pedestrian detection while preserving privacy. Importantly, the obfuscated data remains compatible with publicly available object detectors, requiring no modifications or significant loss in accuracy. Additionally, our technique overcomes the common limitation of relying on labeled sensitive attributes for privacy preservation. By demonstrating the inability of pedestrian attribute recognition models to detect attributes in obfuscated videos, we validate the efficacy of our privacy protection method. Our results suggest that this scalable approach holds promise for enabling camera usage in video analytics while upholding personal privacy.

show abstract

DeepPrivacy2: Towards Realistic Full-Body Anonymization

Cited by 18 publications

References 38 publications

Image Inpainting with Learnable Feature Imputation

Image Inpainting with Learnable Feature Imputation

Computational facial analysis for rare Mendelian disorders

Privacy-preserving visual analysis: training video obfuscation models without sensitive labels

Contact Info

Product

Resources

About