Joao C. A. Almeida inquired about 3D reconstruction software.
 
For those with limited budgets ;-), my lab has found both
NetV and XEVA to be of good value; i.e. they are public domain.
 
We collect data from a Noran Odyssey in NIH Image TIFF format
invert the stack (Macintosh graphics are B/W inverted relative to
unix graphics), save a binary stack to disk and ftp them to a SGI
Indigo for further processing...
 
NetV is a distributed computing model but requires an account
at the San Diego Supercomputer Center (SDSC). All the intensive
calculations are done on a SGI Challenge while your local
workstation handles the easy stuff: classification, definition
of borders, orientation, lighting etc. A link is setup at bootup
so commands to render result in the appearance of images 60-120
seconds later for a 40 x 256 x 256 dataset. This gives good user
feedback despite transmission time. Obviously, render times are
affected by network traffic, high/low resolution or anti-aliasing.
 
The client version of NetV is available for most unix workstations.
For folks whose institutions are members of the SDSC consortium,
contact [log in to unmask] for further info.
 
XEVA is a free-standing application developed by Daniele Marini
at the Universita degli Studi di Milano. It runs on SGI, HP, Sun
and IBM RS6000 machines. Xeva has more pre-processing functions
than NetV. However, XEVA's strongest feature is its open structure
and very well-designed classifier. It is very easy to iteratively
test multiple rendering schemes by linking different classifier
modes to rendering windows. Within any classifier mode, an unlimited
number of LUTs can be set to segment the dataset in a mixed threshold
model (we rarely use more than 10-12 LUTS). Alpha values can be
set for each, and one can magnify the histogram for precise setting
of LUT boundaries and ramps.
 
Orientation of the volume in XEVA is done with user-feedback of a
'virtual volume'. Bascially, a cube with handles shows a
sub-sampled image of your volume so the user can see the major
features moving in real time. On pure Xwindows machines without
hardware acceleration, this can be slower than a SGI with geometry
chips but its still acceptable on non-hardware accelerated systems.
Render times range 3-8 minutes/image on a R3000 Indigo Elan for
a 40 x 256 x 256 dataset. Obviously, shorter render times can
be obtained from newer hardware. In that light, we use NetV for
fast rendering jobs and XEVA for finished products- but optimizing
the classification is best done in XEVA.
 
 
Daniele Marini can be reached at [log in to unmask]
His web page shows some sample images and a link to downloading
XEVA is there;  http://www.dsi.unimi.it/Users/imaging/HOME.html.
 
Parenthetically, neither NetV or XEVA have 3D measurement tools now.
XEVA may implement them soon (an inactive icon suggests this). Finally,
both programs will generate movies - for multispectral datasets
we use SDSC's imcomp (part of the imtools suite of unix image
manipulation tools). imcomp does a digital compositing operation
with alpha channels on one or 2 input images. Output can be in any
of some 40 formats supported by imtools.
 
Daniel Chin
Research Scientist, Agouron Institute
[log in to unmask]