Privacy Beyond Pixels: Latent Anonymization for Privacy-Preserving Video Understanding
Not on the map. Why?
Abstract (arXiv)
We introduce a novel formulation of visual privacy preservation for video foundation models that operates entirely in the latent space. While spatio-temporal features learned by foundation models have deepened general understanding of video content, sharing or storing these extracted visual features for downstream tasks inadvertently reveals sensitive personal information like skin color, gender, or clothing. Current privacy preservation methods focus on input-pixel-level anonymization, which requires retraining the entire utility video model and results in task-specific anonymization, making them unsuitable for recent video foundational models. To address these challenges, we introduce a lightweight Anonymizing Adapter Module (AAM) that removes private information from video features while retaining general task utility. AAM can be applied in a plug-and-play fashion to frozen video encoders, minimizing the computational burden of finetuning and re-extracting features. Our framework employs three newly designed training objectives: (1) a clip-level self-supervised privacy objective to reduce mutual information between static clips, (2) a co-training objective to retain utility across seen tasks, and (3) a latent consistency loss for generalization on unseen tasks. Our extensive evaluations demonstrate a significant 35% reduction in privacy leakage while maintaining near-baseline utility performance across various downstream tasks: Action Recognition (Kinetics400, UCF101, HMDB51), Temporal Action Detection (THUMOS14), and Anomaly Detection (UCF-Crime). We also provide an analysis on anonymization for sensitive temporal attribute recognition. Additionally, we propose new protocols for assessing gender bias in action recognition models, showing that our method effectively mitigates such biases and promotes more equitable video understanding. https://joefioresi718.github.io/SPLAVU_webpage/
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 48 of its 73 references that are in the map. First posted on arXiv 11 Nov 2025.
Builds on
48 of 73 references.
References · 48
- 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
- 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,713
- 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
- 2017Quo Vadis, Action Recognition? A New Model and the Kinetics Datasetcited by 3,270
- 2020Shortcut Learning in Deep Neural Networkscited by 1,969
- 2017Accurate, Large Minibatch SGD: Training ImageNet in 1 Hourcited by 1,492
- 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,432
- 2019Generalized Intersection over Union: A Metric and A Loss for Bounding Box Regressioncited by 1,368
- 2019Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalizationcited by 1,247
- 2017The "something something" video database for learning and evaluating visual common sensecited by 1,103
- 2016NTU RGB+D: A Large Scale Dataset for 3D Human Activity Analysiscited by 686
- 2018Real-world Anomaly Detection in Surveillance Videoscited by 577
36 more
- 2024Revisiting Feature Prediction for Learning Visual Representations from Videocited by 531
- 2023VideoMAE V2: Scaling Video Masked Autoencoders with Dual Maskingcited by 528
- 2022InternVideo: General Video Foundation Models via Generative and Discriminative Learningcited by 447
- 2017Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraintscited by 420
- 2021Differentially Private Fine-tuning of Language Modelscited by 400
- 2023TriDet: Temporal Action Detection with Relative Boundary Modelingcited by 142
- 2021TCLR: Temporal Contrastive Learning for Video Representationcited by 136
- 2019Why Can't I Dance in the Mall? Learning to Mitigate Scene Bias in Action Recognitioncited by 117
- 2020Image Representations Learned With Unsupervised Pre-Training Contain Human-like Biasescited by 107
- 2017Towards a Visual Privacy Advisor: Understanding and Predicting Privacy Risks in Imagescited by 101
- 2022FedPrompt: Communication-Efficient and Privacy Preserving Prompt Tuning in Federated Learningcited by 91
- 2022MGFN: Magnitude-Contrastive Glance-and-Focus Network for Weakly-Supervised Video Anomaly Detectioncited by 89
- 2019Predictive Inequity in Object Detectioncited by 77
- 2018Learning to Anonymize Faces for Privacy Preserving Action Detectioncited by 63
- 2023FACET: Fairness in Computer Vision Evaluation Benchmarkcited by 58
- 2022SPAct: Self-supervised Privacy Preservation for Action Recognitioncited by 47
- 2021Motion-aware Contrastive Video Representation Learning via Foreground-background Mergingcited by 45
- 2021Time-Equivariant Contrastive Video Representation Learningcited by 44
- 2016Privacy-Preserving Human Activity Recognition from Extreme Low Resolutioncited by 43
- 2023STPrivacy: Spatio-Temporal Privacy-Preserving Action Recognitioncited by 29
- 2022Mitigating and Evaluating Static Bias of Action Representations in the Background and the Foregroundcited by 27
- 2021Unsupervised Discriminative Embedding for Sub-Action Learning in Complex Activitiescited by 26
- 2023TeD-SPAD: Temporal Distinctiveness for Self-supervised Privacy-preservation for video Anomaly Detectioncited by 23
- 2021Learning Representational Invariances for Data-Efficient Action Recognitioncited by 22
- 2022PrivHAR: Recognizing Human Actions From Privacy-preserving Lenscited by 16
- 2023Person Re-Identification without Identification via Event Anonymizationcited by 16
- 2022Privacy-Preserving Action Recognition via Motion Difference Quantizationcited by 14
- 2023Preserving Modality Structure Improves Multi-Modal Learningcited by 13
- 2023Tubelet-Contrastive Self-Supervision for Video-Efficient Generalizationcited by 12
- 2020Investigating Bias in Image Classification using Model Explanationscited by 11
- 2023No More Shortcuts: Realizing the Potential of Temporal Self-Supervisioncited by 11
- 2017ConvNets and ImageNet Beyond Accuracy: Understanding Mistakes and Uncovering Biasescited by 7
- 2025Safe-LLaVA: A Privacy-Preserving Vision-Language Dataset and Benchmark for Biometric Safetycited by 7
- 2022Privacy-Preserving Visual Localization with Event Camerascited by 6
- 2025ALBAR: Adversarial Learning approach to mitigate Biases in Action Recognitioncited by 5
- 2022Decouple-and-Sample: Protecting sensitive information in task agnostic data releasecited by 4
Built on by
Citing papers per year: 2026: 1.
1 paper.