ProFace

ProFace: A trustworthy facial data protection research platform developed by Chongqing University of Posts and Telecommunications (CQUPT). It provides efficient implementations of versatile methods for facial data security analysis and privacy protection developed by CQUPT.

This project consists of several modules:

• FacePrivacy: methods for protecting facial privacy in multiple scenarios.
• FaceSecurity: methods for facial data analysis (e.g., DeepFake detection, forensic analysis).
• Platform： A versitile platform for multimedia security analysis and privacy protection, this platform currently includes two functions:face forgery detection and text sentiment analysis.

FacePrivacy

This module implements various algorithms for facial privacy protection.

Paper List:

2025.02: 基于妆容风格补丁激活的对抗性人脸隐私保护 《计算机科学》2025. [code]

2025.01: iFADIT: Invertible Face Anonymization via Disentangled Identity Transform Arxiv. [paper][code]

2024.04: PRO-Face C: Privacy-Preserving Recognition of Obfuscated Face via Feature Compensation IEEE TIFS 2024. [paper][code]

2023.11: Invertible Image Obfuscation for Facial Privacy Protection via Secure Flow IEEE TCSVT 2023. [paper][code]

2022.10: PRO-Face: A Generic Framework for Privacy-preserving Recognizable Obfuscation of Face Images ACM Multimedia 2022. [paper][code]

FaceSecurity

This module implements various algorithms for facial data analysis.

Paper List:

2025.:Deepfake Detection Leveraging Self-Blended Artifacts Guided by Facial Embedding Discrepancy IEEE TCSVT under review.[code]

2024.11: Advancing Generalized Deepfake Detector with Forgery Perception Guidance ACM Multimedia 2024. [paper][code]

2024.08: Inspector for Face Forgery Detection: Defending Against Adversarial Attacks From Coarse to Fine IEEE TIP 2024. [paper][code]

Platform

this module currently includes two functions:face forgery detection and text sentiment analysis.Implementation of the face forgery detection module: When users upload a facial image, the system returns the probability indicating whether the image is forged and a heatmap. Implementation of the text sentiment analysis module: When users enter texts, the system returns the possible emotions contained in the utterance.