Dataset Restricted Access

Replay-Mobile

Costa-Pazo, Artur; Bhattacharjee, Sushil; Vazquez-Fernandez, Esteban; Marcel, Sébastien

Replay-Mobile is a dataset for face recognition and presentation attack detection (anti-spoofing). The dataset consists of 1190 video clips of photo and video presentation attacks (spoofing attacks) to 40 clients, under different lighting conditions. These videos were recorded with an iPad Mini2 (running iOS) and a LG-G4 smartphone (running Android).

 

Database Description

All videos have been captured using the front-camera of the mobile device (tablet or phone). The front-camera produces colour videos with a resolution of 720 pixels (width) by 1280 pixels (height) and saved in ".mov" file-format. The frame rate is about 25 Hz. Real-accesses have been performed by the genuine user (presenting one's true face to the device). Attack-accesses have been performed by displaying a photo or a video recording of the attacked client, for at least 10 seconds.

Real client accesses have been recorded under five different lighting conditions (controlled, adverse, direct, lateral and diffuse). In addition, to produce the attacks, high-resolution photos and videos from each client were taken under conditions similar to those in their authentication sessions (lighton, lightoff).

The 1190 real-accesses and attacks videos were then grouped in the following way:

  • Training set: contains 120 real-accesses and 192 attacks under different lighting conditions;
  • Development set: contains 160 real-accesses and 256 attacks under different lighting conditions;
  • Test set: contains 110 real-accesses and 192 attacks under different lighting conditions;
  • Enrollment set: contains 160 real-accesses under different lighting conditions, to be used **exclusively** for studying the baseline performance of face recognition systems. (This set is again partitioned into 'Training', 'Development' and 'Test' sets.)

 

Attacks

For photos attacks a Nikon coolix P520 camera, which records 18Mpixel photographs, has been used. Video attacks were captured using the back-camera of a smartphone LG-G4, which records 1080p FHD video clips using its 16 Mpixel camera.
 

Attacks have been performed in two ways:

  1.  A matte-screen was used to perform the attacks (i.e., to display the digital photo or video of the attacked identity). For all such (matte-screen) attacks, a stand was used to hold capturing devices.
  2. Print attacks. For "fixed" attacks, both capturing devices were supported on a stand (as for matte-screen attacks). For "hand" attacks, the spoofer held the capturing device in his/her own hands while the spoof-resource (printed photo) was stationary.

In total, 16 attack videos were registered for each client, 8 for each of the attacking modes described above.

  • 4 x mobile attacks using an Philips 227ELH screen (with resolution 1920x1080 pixels)
  • 4 x tablet attacks using an Philips 227ELH screen (with resolution 1920x1080 pixels)
  • 2 x mobile attacks using hard-copy print attacks fixed (produced on a Konica Minolta ineo+ 224e  color laser printer) occupying the whole available printing surface on A4 paper
  • 2 x mobile attacks using hard-copy print attacks fixed (produced on a Konica Minolta ineo+ 224e  color laser printer) occupying the whole available printing surface on A4 paper
  • 2 x mobile attacks using hard-copy print attacks hand (produced on a Konica Minolta ineo+ 224e  color laser printer) occupying the whole available printing surface on A4 paper
  • 2 x mobile attacks using hard-copy print attacks hand (produced on a Konica Minolta ineo+ 224e  color laser printer) occupying the whole available printing surface on A4 paper

 

Reference

If you use this database, please cite the following publication:

Artur Costa-Pazo, Sushil Bhattacharjee, Esteban Vazquez-Fernandez and Sébastien Marcel,"The REPLAY-MOBILE Face Presentation-Attack Database", IEEE BIOSIG 2016.
10.1109/BIOSIG.2016.7736936
http://publications.idiap.ch/index.php/publications/show/3477

Restricted Access

You may request access to the files in this upload, provided that you fulfil the conditions below. The decision whether to grant/deny access is solely under the responsibility of the record owner.


Access to the dataset is based on an End-User License Agreement. The use of the dataset is strictly restricted to non-commercial research.

Please provide us the following information about the authorized signatory (MUST hold a permanent position):

  • Full name
  • Name of organization
  • Position / job title
  • Academic / professional email address
  • URL where we can verify the information details

The requester must use their personal valid email address from the same organization as the signatory to request access.


1,557
260
views
downloads
Views 1,557
Downloads 260
Data volume 3.0 TB
Unique views 1,152
Unique downloads 148

Share

Cite as