HD_UHD_Eyetracking_Videos 37 videos viewed in HD or in UHD with eyetracking data
|Created the||28 Jan 2016|
Presentation & Goal
UHD TV standard defines new video technologies as an increasing resolution from HD (1920x1080) to UHD, i.e. 4K (3840x2160). The emergence of UHD potentially provides a better immersion of the user thanks to a wider visual angle with appropriate larger screens. Indeed, ITU defines the optimal viewing distance as the distance at which scanning lines just cannot be perceived with visual acuity of 1'. It is thus set to 3H for HD and 1.5H for 4K-UHD where H is the height of the screen. This increase of resolution and stimulated visual angle can modify visual attention deployment and visual patterns of people looking at HD and UHD videos.
The experiment was conducted in a test environment set as a standard subjective quality test condition according to ITU-R BT.500. The HD display used was a 46'' Panasonic Full HD Vieta and the 4K display used was a 65'' Panasonic TX-L65WT600E. The viewing distance was 1.5H, i.e. 120 cm, in UHD and 3H, i.e. 170 cm, as recommended in ITU-R BT.1769.
In HD, 34 naive observers (17 males and 17 females) participated in this experiment and in UHD, 36 naive observers (18 males and 18 females). UHD and HD were assessed in two different sessions with different observers to avoid any effect of memorization.
We adopted a free-looking approach in these experiments. Sequences were randomized for each observer. They were 2 seconds spaced out. The whole test lasted approximately 25 minutes.
The dataset is composed of 37 native UHD high quality video sequences from seven content provider. In HD, the original sequences were downscaled with Lanczos-3 algorithm.
For each video sequence, gaze point, fixation and saccade data are provided.
Because of the larger stimulated visual angle in UHD, observers can need to move more their head and eye tracking systems may not be accurate enough at the edges of the screen. We developed a new setup to address this issue: the Eye Head Tracker (EHT). EHT is a combination of the mobile SMI eye tracking glasses and of the head tracker OptiTrack ARENA.
We implemented an application which collects these two data in order to provide the gaze position in the screen plane.
The main goal of this dataset is the comparison of visual attention and viewing behavior in HD and UHD.Different kind of analyses can be done: comparison of saliency through fixation density maps, impact of viewing conditions and resolution on distribution of gaze points and fixations, comparison of distribution of saccades, etc. Moreover, this dataset can be used to evaluate the performance of visual saliency models in HD and UHD, by comparing fixation density maps computed from acquired data with simulated saliency maps.
Futhermore, this dataset provides useful data for any researcher working on dynamic visual attention in videos (dynamic visual attention modelling, visual attention and quality of experience, saliency-based video compression, etc.).
The main qualities of the dataset are the large number of sources and observers compared to previously published video saliency database, as well as the high quality of professional videos.
Heat map of the same video in UHD and in HD
Source : HD and UHD videos
The dataset is composed of 37 native UHD high quality video sequences from seven content provider: SJTU Media Lab, Big Bug Bunny (Peach open movie project), Ultra Video Group, Elemental Technologies , Sveriges Television AB (SVT), Harmonic, Tears of steel (Mango open movie project).
In HD, the original sequences were downscaled with Lanczos-3 algorithm.
The frame rates of the original sequences varies from 25 to 120 fps. They were uniformly played frame by frame with 25 fps in our test, causing some movements to appear a bit slower than in reality.
Each source was cut to 12 second long clips at 25 fps, producing a total of 300 frames each.
Result : Gaze points
For each video and each observer, the following gaze data are stored: eye identifier (0 for left eye and 1 for right eye) ; time (sec) ; eye position in X axis (px) ; eye position in Y axis (px). The origin (0,0) is in the upper left corner of the frame. If the eye was not tracked by the eye tracker, the X and Y positions are set as NaN.
The mean of successive left and right eye positions might be calculated to obtain binocular information.
Result : Fixations and saccades
A fixation is defined as the status of a region centered around a pixel position which was stared at for a predefined duration. A saccade corresponds to the eye movement from one fixation to another.
For each source, we provide the following data about fixations: starting time of fixation (ms) ; end of fixation (ms) ; fixation position in X axis (px); fixation position in Y axis (px); number of gaze points in the fixation; observer number.
We also provide saccade data between fixations as follows: starting time of saccade (ms) ; end of saccade (ms); position of start of saccade in X axis (px); position of start of saccade in Y axis (px) ; position of end of saccade in X axis (px) ; position of end of saccade in Y axis (px) ; saccade length (px) ; saccade orientation (°) ; observer number.
|Video resolution||1920x1080 or 3840x2160|
|Video length||250 to 300 frames|
|Video duration||10 to 12 seconds|
|Number of videos||37|
|Observation distance||3H (HD) or 1.5H (UHD)|
|Pre screening||Snellen, Ishihara|
|Number of observers||42 naives (8 rejected) for HD and 40 (4 rejected) for UHD|
|Age (Mean [Range])||24 [19 44] years for HD and 28 [19 56] for UHD|
|Male / Female repartition||19 / 23 for HD and 19 / 21 for UHD|
|Display||Panasonic 3D Full HD Vieta (1020 x 573 mm) for HD and Panasonic LCD TV TX-L65WT600E (1428 x 804 mm) for UHD|
|Display resolution||1920x1080 and 3840x2160|
|Model||SMI ETG + EyeHeadTracker|
|Acquisition frequency||30Hz Binocular|
|Video name||Preview||SI||TI||Number Frame||Native Frame Rate||Played Frame Rate|
Database Download Guide for ftp://ftp.ivc.polytech.univ-nantes.fr/IRCCYN_IVC_HD_UHD_Eyetracking_Videos/
Use a FTP Manager to download all files at once
With Filezilla (Windows, Linux, Mac OS)
- Download the FileZilla Client software at https://filezilla-project.org/ and install it.
- Launch File > Sites Manager and fill in the form. The Host is ftp.ivc.polytech.univ-nantes.fr. Use Anonymous for Logon Type. All others fields can be left blank. Finally, press Connect.
- In the first panel, you can see connection's logs. Normally, a message like Response: 220 ftproxy FTP server (Version 184.108.40.206 - 2005/01/11 13:03:28) ready. is printed.
- The left column below contains your computer file system and the selected folder. In the right column, you can see the FTP server file system and its selected folder.
- In the left colum, select the local directory in which receive the database. In the right panel, find IRCCYN_IVC_HD_UHD_Eyetracking_Videos/ , right click and press Download.
- Bottom panel shows file transfer informations. All files of the databases are being downloaded in your computer. Please wait during the transfer.
With WGET (Linux)
Copy and paste this command in a terminal :
wget -P your/destination/folder ftp://ftp.ivc.polytech.univ-nantes.fr/IRCCYN_IVC_HD_UHD_Eyetracking_Videos/