--- title: "Printed 3D Masks Attacks Dataset" description: "3,800+ videos 5 devices" url: "https://unidata.pro/datasets/printed-3d-masks-attacks/" date_modified: "2025-12-09T10:19:22+03:00" language: "en-US" --- It is a diverse 3D mask attacks dataset containing over 3,800 videos of individuals wearing or holding 3D printed face masks, designed for training facial recognition and liveness detection models, with detailed metadata and realistic mask attacks to support robust anti-spoofing systems and image recognition tasks. ## Dataset Structure ### The Numbers Section **Numbered list:** - **Number:** 3,800+ — **Text:** videos - **Number:** 5 — **Text:** devices ### Tooltips Section **Tooltip items:** - **Name:** iBeta - **Name:** Liveness Detection - **Name:** Computer Vision - **Name:** Security ### Dataset Information **Table with data:** | Characteristic | Data | | --- | --- | | Description | Videos of individuals wearing or holding 3D masks | | Data types | Video | | Tasks | Liveness Detection, Computer Vision, iBeta | | Number of video | 3 800+ | | Labeling | Metadata (age, gender, ethnicity, devices) | | Gender | Male, Female | | Number of attributes | 31 | **Media Slider:** - **Video on Slayder:** - **Video on Slayder:** **Link to the sample:** [Download sample](https://drive.google.com/drive/folders/1G6NDwA0efI0NoOJuEt5WfXzLHuN_io9c) ### Statistics - Charts **Charts with Titles:** - **Shortcode:** [ays_chart id="39"] — **caption above the graph:** Age of the unique people - **Shortcode:** [ays_chart id='38'] — **caption above the graph:** Gender distribution - **Shortcode:** [ays_chart id="40"] — **caption above the graph:** Ethnicity of the unique people ### Technical Specifications **Table with data:** | Characteristic | Data | | --- | --- | | Video extension | MP4, MOV | | Video Resolutions | Min = 1920х1080, Max = 3840х2160 | | Video duration | 4 second | | Number of background | 9 | | Video overlap | No more than 5% | | Devices | IOS, Android | **Source and data collection methodology:** Source and collection methodology. Data was collected by a partner of Unidata. ### Dataset Use Cases - Slider **Industry Cards:** - **Industry:** Biometrics & Security Systems — **Title:** Improving Facial Recognition Anti-Spoofing Models — **Text:** Printed 3D Masks Attacks Dataset provides a reliable foundation for developing anti-spoofing algorithms in facial recognition systems. The dataset contains high-resolution face images and recordings of mask attacks using 3D prints and replica face masks. By analyzing these samples, researchers can train image recognition models to differentiate between real faces and 3D mask attacks, improving the robustness of biometric authentication and identity verification technologies. - **Industry:** Artificial Intelligence & Machine Learning — **Title:** Training Models for Mask Attack Detection — **Text:** This dataset serves as high-quality training data for machine learning and deep learning systems designed to detect spoofing attempts. It comprises both genuine and printed 3D models, offering balanced samples for supervised learning. With structured data collection and labeled segmentation masks, it supports the development of algorithms capable of recognizing subtle texture and lighting differences between face masks and real human skin. - **Industry:** Digital Forensics & Fraud Prevention — **Title:** Enhancing Security in Identity Verification Systems — **Text:** In digital forensics, Printed 3D Masks Dataset aids experts in analyzing mask attacks used to deceive facial recognition systems. The dataset’s diversity of 3D prints and video masks dataset samples enables the testing of image segmentation techniques to isolate mask regions from authentic face images. Such data improves fraud detection systems, strengthening quality control in biometric verification and reducing false acceptances in high-security applications. - **Industry:** Computer Vision Research & Academia — **Title:** Image Recognition Models Against Spoofing Attempts — **Text:** For academic research and computer vision studies, this Printed 3D Masks Attacks Dataset offers valuable resources for testing and benchmarking image recognition models under spoofing conditions. The dataset consists of diverse samples of face masks, 3D models, and segmentation masks, allowing the evaluation of model accuracy in detecting artificial reproductions. It supports innovation in facial recognition, image segmentation, and deep learning architectures, contributing to more secure and trustworthy biometric technologies. ### Fact **List of Questions:** - **Question:** Is it possible to request a custom dataset? — **Answer:** Yes. You can request a customized dataset tailored to your research needs, including specific mask materials, facial angles, lighting conditions, or device types. Unidata can also generate custom 3D mask attack videos with defined demographic attributes for enhanced testing in biometric and security applications. - **Question:** How was the data collected? — **Answer:** All videos were collected by a verified Unidata partner under standardized conditions using iOS and Android devices. The recording process ensures diversity across backgrounds, resolutions, and camera types while maintaining strict privacy controls and data collection ethics. - **Question:** How are Unidata datasets licensed? — **Answer:** Unidata datasets follow a dual-licensing model: free samples are available for testing, while full datasets can be purchased for commercial use. - **Question:** Do Unidata datasets follow GDPR or other data privacy regulations? — **Answer:** Yes. All Unidata datasets are curated in full compliance with GDPR and relevant international data protection laws. The data is collected exclusively from legally authorized sources, ensuring that every video and metadata record adheres to ethical and lawful usage standards. - **Question:** How are Unidata datasets stored? — **Answer:** All datasets are securely hosted on AWS cloud infrastructure, offering high availability and scalability for large-scale data handling. Unidata’s storage and management practices comply with ISO 27001 and ISO 27701 certifications, providing a secure, privacy-oriented framework for sensitive video and biometric data. - **Question:** How long does it take to receive the dataset? — **Answer:** After you submit your request, Unidata will contact you to verify details and finalize documentation. Once agreements are signed and payment is confirmed, the dataset is delivered within 3–10 business days. - **Question:** Is this a real-world dataset or synthetic data? — **Answer:** This is a real-world dataset composed of authentic videos of individuals interacting with 3D-printed face masks. The data was recorded using real devices in controlled environments to simulate realistic mask attacks, making it ideal for testing the resilience of facial recognition technologies. - **Question:** Why are 3D printed masks more difficult for biometric systems to detect? — **Answer:** Unlike flat printed photos, 3D masks reproduce facial depth and geometry, making them more challenging presentation attacks for facial recognition systems. This dataset helps AI models learn subtle visual cues that distinguish real faces from realistic 3D mask attacks. [Full list of this site's AI-readable pages](https://unidata.pro/llms.txt)