Light Field Video Sequence by Nagoya University
This
test sequence (Tunnel_Train_2) is provided by
Fujii Laboratory
Department of Information and
Communication Engineering
Graduate
School of Engineering, Nagoya University
Term of Use: ANY kind of publication or report using this sequence MUST refer to the following TWO references.
[1]
http://www.fujii.nuee.nagoya-u.ac.jp/multiview-data/
[2] Mehrdad
Panahpour Tehrani, Sho Mikawa,
Shu Fujita, Toshiaki Fujii: g[MPEG-I Visual]
Introduction to A New Test Sequence gTunnel_Train_2h
Captured by Light Field Video Camerah
ISO/IEC
JTC1/SC29/WG11 MPEG2018/ m41995, Gwangju, Korea, January 2018.
Copyright: ONLY Available for Academic Usage
Production: Nagoya University
Capturing Condition:
We have placed objects close to the camera as shown in Fig. 1
(right), so the disparity among views is distinguishable.
The detail of the capturing condition is in Fig. 1 (left).
The moving blue-doll is at collimation plane that should have
disparity equal to 0 (about 40 cm from camera).
The nearest object is placed at about 40 cm from the camera.
Fig. 1. Capturing (left) scene, and (right) condition.
Specification of the Camera and the Main Lens:
Here, we introduce the specification of the focused plenoptic camera, and the main lens, used for the
capturing.
Fig. 2 shows the camera and lens used in this capturing.
Tables 1 and 2 show the specifications of the Raytrix [2, 3] camera and the main
lens, LMVZ166HC (by Kowa) [7],
respectively.
The setting of the main lens is: focal length = 16mm, F-stop:
F2.8.
Fig. 2. Camera (Raytrix R5-C-
GigE-F2.4 (color)), and lens (LMVZ166HC (by Kowa)) used for capturing.
Table 1. Raytrix specification.
Product
code (model number) |
R5-C- GigE-F2.4 |
Resolution |
2048 (H) x
2048 (V) 4 million pixels |
Image
sensor |
Progressive
scan CMOS 1 inch CMOSIS CMV 4000 |
Shutter |
Global
shutter |
Pixel size |
5.5 Κm ~ 5.5 Κm |
Frame Rate
(Raw) |
25 fps @
Dual GigE |
Resolution after
reconstruction |
Approximately
1 million pixels |
Rx microlens |
F 2.4 |
Number of
pixels @ microlens diameter |
23 pixels /
1 microlens |
Number of
depth layers |
Approximately
90 layers |
Depth of
field |
Approximately
6 times the usual lens |
Lens mount |
C mount |
Output
interface |
Dual GigE |
Viewer
software |
RxLive software (refocus, all focus, 3D, multi view, stereo display) |
SDK |
4D Light
Field SDK (separately) |
Usage
environment |
MS - Windows
7, CUDA, OpenGL 3.0, Compute Capability |
Onboard GPU
|
Nvidia CUDA GPU GeForce GTX-980 or higher recommended |
Power
supply |
Power
supply from external power supply |
Table 2. Main lens specification
Focal Length |
16 - 64mm (4x) |
Focal Length Sort Order |
016 |
Lens Type |
Varifocal |
Image Size |
1" (12.8 x 9.6 x
16mm) |
Iris Range (F-Stop) |
F1.8 – 16 |
Focusing Range |
1.0m |
Filter Thread Size |
M58x0.75 |
Mount |
C-mount |
Test Sequence:
Below, in Figures 3, 4 and 5, we show the captured image (raw: Bayer color filter array & color)
at frame 90 and with different magnifications.
Multiview data is generate using the method explained in [1].
Note that the captured data has 25fps, while the arranged data is
set to 30fps.
The parameters of the captured data are as follows:
Raw Video Data
·
Resolution: 2048x2048 pixels
·
Color: 8 bits, PNG
·
Frame rate: 30 fps
·
Number of frames: 300
·
Multiview Video of Raw Data
·
Viewpoints: 5 x 5 (25 views points, 2D arrangement)
·
Resolution per view: 920x880 pixels
·
Color: 8 bits, PNG
·
Frame rate: 30 fps
·
Number of frames: 300
Color
Video Data
·
Resolution: 2048x2048 pixels
·
Color: 24 bits, PNG
·
Frame rate: 30 fps
·
Number of frames: 300
Multiview Video of Color Data
·
Viewpoints: 5 x 5 (25 views points, 2D arrangement)
·
Resolution per view: 920x880 pixels
·
Color: 24 bits, PNG
·
Frame rate: 30 fps
·
Number of frames: 300
Camera parameters that are given
by (SDK output)
·
Distance from
the center of the image to the center of the central microlens
·
Diameter, and rotation angle of microlens,
and lens array, respectively.
·
Depth
information for the three type of the mircolenses
·
Distance between different types of microlenses
·
The thickness of the microlens
boundary line
·
Lenses
Distortions
Fig. 3. Captured
raw image at frame 90 by Raytrix, R5-C- GigE-F2.4.
Fig. 4. Magnified region of an area in Fig. 3.
Fig. 5. Magnified
region of an area in Fig. 4
Test Data:
Lenslet: Download Raw
Multiview: Download Raw Multiview
Lenslet: Download Color
Multiview: Download Color Multiview
Raytrix SDK output, xml: Download Camera Parameter
Converter:
Lenslet to Multiview Conversion Tool (executable): Download
Note: This executable tool is only tested on Windows 10.
Reference
[1] Mehrdad Panahpour Tehrani, Sho Mikawa, Yuto Kobayashi, Shu
Fujita, Keita Takahashi, Toshiaki Fujii: g[MPEG-I Visual]
Development of a 3D Imaging System Using Light Field Camera and Tensor
Displayh, ISO/IEC JTC1/SC29/WG11 MPEG2017/ / M41244, July 2017, Torino, Italy.
[3] Christian Perwaß and Lennart Wietzke: gSingle Lens 3D-Camera with Extended
Depth-of-Fieldh, proceedings of
SPIE - The International Society for Optical Engineering 8291:4- · February
2012.
[4] Gordon Wetzstein, Douglas Lanman, Matthew Hirsch, Ramesh Raskar:
gTensor Displays: compressive light field synthesis using multilayer displays
with directional backlightingh, TOG, vol. 31, no. 4, 2012.
[5] Ren Ng, Marc Levoy, Mathieu Bredif, Gene Duval, Mark Horowitz, Pat Hanrahan: gLight
Field Photography with a Handheld Plenoptic Camerah,
Stanford University Computer Science Tech Report CSTR, vol. 2, no. 11, 2005.
[6] Todor Georgiev,
and Andrew Lumsdaine: gFocused Plenoptic
Camera and Renderingh, Journal of Electronic Imaging, vol. 19, no 2, 2010.
[7] http://www.kowa-optical.co.jp/fa/e/
Acknowledgment
This
work is partially supported by Grant-in-Aid for Scientific Research (C)
registered number 16K06349.