Datasets

Multimedia Forensics

Perception of Synthetic Faces

We release the dataset and collected data of the work:


F. Lago, C. Pasquini, R. Böhme, H. Dumont, V. Goffaux, G. Boato, "More Real than Real: A Study on Human Visual Perception of Synthetic Faces", accepted for publication in IEEE Signal Processing Magazine (2021).


The material can be downloaded here.

FF++ Social

FF++ Social contains videos from the widely known dataset FaceForensics++ (validation and testing splits) that have been shared through Facebook and YouTube. The dataset is intended to support researchers in multimedia forensics in evaluating the generalization ability of forensic detectors when dealing with data circulating on the web through popular sharing platforms.

Download

The FF++ Social dataset has been used in the work:

F. Marcon, C. Pasquini, G. Boato, "Detection of manipulated face videos over social networks:

a large-scale study", submitted to Journal of Imaging (under minor revision).

RAISE

RAISE is a challenging real-world image dataset, primarily designed for the evaluation of digital forgery detection algorithms.

Website

ISIMA

The ISIMA (Images Sharing via Instant Messaging App) dataset contains images that are shared between Android phone and iOS phone via three instant messaging applications Facebook Messenger, Whatsapp, and Telegram. All images are stored in JPEG format

Website

RSMUD & VSMUD

To facilitate research on tracking social network origin of images, we collect two datasets: R-SMUD (RAISE Social Multiple Up-Download), and V-SMUD (VISION Social Multiple Up-Download).

All images are shared at maximum three times through three platforms: Facebook (FB), Flickr (FL), Twitter (TW).

Website

DECEPTION

Website

Computer Vision and Behaviour Analysis

Re-DID

All the videos in the dataset are retrieved from YouTube; 25 of them are recorded using car mounted Dash-Cams, the remaining ones have been taken by other devices such as mobile phones.

Website

UCD

The UNITN Crowd Dataset consists of two video sequences which have been segmented into two different sub-sequences each, and used for both crowd motion segmentation and anomaly detection.

Website

USID

The UNITN Social Interaction(USI) Dataset consists of 4 types of two-person interactions: Talking, Shaking, Hugging and Fighting. Each type of two-person interaction has 16 samples, with the total number of 16x4 = 64 samples.

Website

Semantic Retrieval

ACM EiMM DB

Website

EventMask

The EventMask Dataset consists of two archives of images. Each archive contains three directories, one with the original images used for the game and another one with the event saliency maps already in binary form.

Website

MediaEvalSEM

year 2015

year 2014