Quasi3D Public Outputs
These are the public outputs from the EPSRC grant of the above name.
This project and its related predecessor were supported by EPSRC. The
papers are shown in increasing date order.
-
"An Efficient 2.5D Rendering and Compositing System"
Max Froumentin and Philip Willis,
Computer Graphics Forum 18 (3) 1999 Eurographics 99 Conference issue
pp. C385-C394 and C428 ISSN 0167-7055
-
"A Vector-based Representation for Image Warping"
Max Froumentin, Frederic Labrosse and Philip Willis,
Computer Graphics Forum 19 (3) 2000 Eurographics 2000 Conference issue
pp. C385-C394 and C428 ISSN 0167-7055
-
"Demosaicing of Colour Images Using Pixel Level Data-Dependent
Triangulation"
Dan Su and Philip Willis,
Eurogaphics UK Conference
"Theory and Practice of Computer Science,"
3-5 June 2003,
ISBN:0-7695-1942-3, (published by IEEE).
-
"Quasi-3D Cel-based Animation"
Man Qi and Philip Willis,
Proceedings of "Vision, Video and Graphics 2003"
(published by Eurographics)(supported by EG/IMA/SIGGRAPH July 2003),
pp. 111-116, ISBN 3-905673-54-1.
-
"Image Interpolation by Pixel Level Data-Dependent Triangulation"
Dan Su and Philip Willis,
(accepted for Computer Graphics Forum)
-
"Texture Synthesis, Texture Transfer and Constrained Texture Synthesis using
Particle Swarm Optimisation"
Y Zhang, Y, Meng, D Su, P J Willis,
(in preparation).
-
"Image Meshes for CMYK Printing"
Dan Su and Philip Willis (in advanced preparation).
-
"Pixel Level Data Dependent Triangulation with its Applications"
Dan Su's PhD thesis in full (Winzipped PostScript)
-
Dr Man Qi's Summary Technical Report (Microsoft Word document).
(Not published: project use only.)
Quasi3D Examples
This section includes a cross-section of working material used by the project.
They are samples of movie-specification directories, used by us to compile
stacks of cels and animate them over time. In selected cases we have included
avi movies. We also show small jpeg images taken from different
movie frames. There are files showing the language used to locate and
animate the various cels: these would be used to build complete stack files for
each frame, which our renderer then turns into pictures. The rendering output
resolution is chosen independently of the input pictures'. Supersampling allows
us to tailor the output quality. Focus can be adjusted. Lighting can be
dynamically changed. The viewpoint can be dynamically changed.
The sample pictures and files
are to give the reader an idea of the range of tests we performed when
validating the system, as well as give a feel for how it works and the range of
input visual quality supported. The movies are compressed with WinZip, for
faster web transmission.
A note on the quality of the movies: The movies are prepared using Adobe
Premier and we view them with the Microsoft Media Player. Our experience is
that this sometimes introduce defects, notably small runs of
"missing" pixels and ghosting of earlier frames, which are not present in our
source material. Each movie is created from Targa images (lossless) generated
by Quasi3D. As movie files are intrinsically very large, we have restricted the
image size, the frame rate, or both.
-
Snooker table in action. These are high quality graphical (i.e. synthetic)
images such as might be used for animation.
-
Snooker movie
(The movie is rendered in slow motion, partly to demonstrate we can do it but
mainly so that the action can be seen easily.)
-
Lighting animation, going from day to night. This test uses a low quality
digital photograph, with the frontlighting darkening steadily and backlighting
for the windows becoming revealed in the darkness.
-
Garden composited from separate elements. Each element is a mid-quality digital
photograph. The viewpoint varies to illustrate parallax being applied to "real"
images, such as might be used at higher resolution for film work. Each element:
the bacground, the grass and the two statues are on different cels. Note how
the correct movement of these gives a real feel of 3D parallax.
-
Garden movie
(This movie has been given the smoothest animation rate, to show full frame rate.)
-
This scene is based on a low quality (over-magnified) jpeg photograph with
an overlayed moving balloon. It was a quick test to verify both rotational
change of viewpoint and dynamic change of focus. (No movie)
-
A test showing a cel being translated and magnified, composited with a
background scene of a lake. The effect is that the balloon appears to fly
towards the viewer. In addition, we have animated the overall lighting so the
scene moves rapidly to dusk.
-
A computer game-like maze, in which the exit is identified by a flower.
This tests texture maps and dynamically changing viewpoints. In contrast the
the earlier examples, it is the viewer moving in 3D space through fixed images.
In traditional cel animation, the images would have to be redrawn.
(No movie)
-
A direct illustration of our mesh images, used here to zoom. The movie
starts at the native pixel resolution of the image, and then the viewpoint
slowly moves much closer. Note how the detail zooms very smoothly, clearly
showing the benefits of our mesh images.
Phil Willis (PI)
Man Qi (RA)
Dan Su (RS)
