Database contains Kinect-2 recordings of various objects from different perspectives (360 deg). Corresponding RGB and Depth information is available for each object and angle. Database is used for reconstructing 3D model of an object from the point cloud of depth information.

Request Form

Sequential Emotion Database provides videos of people as they changed their emotions, in order to capture the transitions between the different facial muscle states was created. We used four basic emotions as defined by Paul Ekman, including the neutral state at the beginning of each video. We didn't record the emotions for fear and disgust as they are harder to properly mimic.

This database is strictly comprised of natural posed faces where the subjects had no prior knowledge of how to perform certain expressions, nor were they allowed to practice in front of a mirror. Each of the emotions were recorded for 3 seconds, after which the subjects were notified to move on to the next expression. As each sequence of emotions comprised of 5 emotional states, the resulting videos were all around 15.5 sec long. For the sake of redundancy, 5-7 shots of both sequences were recorded for each person and only the best 5 were kept.

Request Form

We have developed a new huge facial expression database called iCV Multi-Emotion Facial Expression Dataset (iCV-MEFED), designed for multi-emotion recognition. The database includes 31250 facial expressions with different emotions of 125 subjects whose gender distribution is almost uniform. Each subject acts 50 different emotions and for each of these emotions 5 samples have been taken by Canon 60D camera under uniform lightening condition with relatively uniform background and resolution of 5184x3456 per sample.

All emotion expressions in iCV MEFED database are presented in complimentary-dominant format. We used 7 basic emotional states :  1-Anger, 2-Contempt, 3-Disgust, 4-Fear, 5-Happiness, 6-Sad, 7-Surprise, N-Neutral. Each image in database is labeled in following format : A_B, where A and B correspond to complimentary and dominant emotion respectively (e.g label 1_7 would mean angrily surprised).

The images are taken and labelled under supervision of psychologists and the subjects have been trained about the acting emotions that they posed.

Request Form

We have compiled a 3D head pose database of 50 subjects with varying poses and respective labels, called SASE. The data has been collected using Kinect2 from 50 subjects. The subjects changed their head-poses in front of the sensor and the captured sequences consist of RGB frames, depth frames and corresponding labels. The distance from the camera fluctuates at around 1m, background is white, and light blue markers were attached to the faces of the subjects.

For each image in database is provided information about coordinates in meters of location of nose and angle values of head pose. Since Kinect produces a certain error and the data was labelled automatically there can be some noise in our data.

Request Form

The SASE-FE dataset contains video set of 50 subjects. For each subject, there are 12 videos representing 6 basic emotions (Anger, Happiness, Sadness, Disgust, Contempt, Surprise) for real and fake expression. Each video was recorded with a high resolution camera with 100 frames per second and it's about 3-4 seconds. In order for the subjects to express these emotions, they were shown videos which are meant to induce these emotions and were acted accordingly.  For the real emotion set subjects were supposed to express the same emotion which was provoked by the shown video. In the second case the expressed emotion and stimulated emotion were contrasted (e.g to record a faked surprise we've shown a calling disgust video and asked to act surprise) .

In each video, subjects started from a neutral emotion and the length of this neutral emotion is not predefined. The sequence of returning to the neutral state was introduced to depict a variation from the acted emotion to the fake emotion to be used in the training process.

Request Form