Dear DAVID Community,

In 2016, DAVID and this forum have been integrated into HP Inc.
This forum has been read-only since then, and will be switched off by the end of 2019.

To everyone who has contributed here, we thank you for all your work and for the constructive, helpful and friendly atmosphere you have created!
Please continue to discuss on the HP Support Forums.

minimum computer system reqirements

The place for all topics related to the 'Shapefusion' tool of DAVID.

minimum computer system reqirements

Postby MrPete » Wed Jun 04, 2008 2:05 am

I've been having computer issues trying to run shapefusion, my PC it was on died after scanning, before fusing files. Then I tried running it on a macbook and as the file size grows the response speed slows to a crawl to the point of going nowhere. Replaced the dead PC with a new one - Dell Vostro 200 with 1GB ram and integrated video (no accelerated video card). The new PC "Dell" can call up some combined files that were done on the macbook but go to do any kind of manipulation such as spin, save as, etc get and unexpected error.

The scans were done at 1600 x 1200 and there are 40 to combine so the files are getting rather large. So large file size is a contributor.

Here are my questions:

Is there a way of eliminating overlapped points from combining multiple scan objects to minimize file size?

Saving in different fashions doesn't seem to effect file size, am I doing something wrong?

I've read the unexpected error is related to video or main memory, is there a minimum required to run this reasonably fast or even just run?

Which will have a bigger effect on getting this running, system RAM or video acceleration/memory?
Attachments
fusion question1.jpg
User avatar
MrPete
 
Posts: 31
Joined: Mon Mar 24, 2008 1:17 am

Postby MagWeb » Wed Jun 04, 2008 7:56 am

Hi, good morning, may you have a high resoluted day

MrPete wrote:The scans were done at 1600 x 1200 and there are 40 to combine so the files are getting rather large.

If five scans are resulting a about 92 MB file and you need 40 scans (do you realy need 40?) the result will be a 736MB final file!!! How do you want to use this giant file?
You have to reduce the amonut of used vertices/faces. You can use Meshlabs quadric edge collapse decimation filter (no, I´m not from Pisa). Other 3d packages Blender/Maya... own such filters too....
If you´re going to stay at this filesize you´ll need real giant system requirements too


Is there a way of eliminating overlapped points from combining multiple scan objects to minimize file size?
No, I don´t know such a possibility. Only the reduction of the whole file.

Saving in different fashions doesn't seem to effect file size, am I doing something wrong?
I think allowing removal in simple fusion would have some effect. -- Did you succeed fusing in simple mode at a res of 600 ?

Loosing details in the final fusing step may be caused by slight inaccuracies of the meshes alignment (or of the meshes themselves). The poisson fusion seems to detect hard edges better than the simple methode

I made the best experiences NOT to combine the meshes BUT to hide all I don't need for one process. So Shapfusion gets a precise target (not doubled data)
So if I have 4 meshes (A, B, C, D) that build a 360° scan, I load all of them.
Now I hide C and D and align A to B;
I hide A, show C and allign C to B;
Now I hide B.
For D is the last scan of the circle it has to fill the gab between A and C (opposite to B)
I show C and D and align D to C
I hide D
Now I show A and C and combine only these two!
I show D and align it to the combined A/C-mesh using only fine registration

Hope that helps
Gunter
MagWeb
Moderator
 
Posts: 2499
Joined: Wed Jul 18, 2007 8:48 pm

Postby leo » Wed Jun 04, 2008 9:52 am

Hi,

Some comment about vostro 200 and the integrate video.

I have a Vostro 200 with 2 GB. Without graphics card was extremely slow working in 3D (davids 3D and Meshlab) I put a Nvidia 9600 card and now works fine. Theoretically these card needs a minimum of 400 Watt power supply and Vostro has only 300 Watts – but it works OK.

Leo
leo
 
Posts: 48
Joined: Fri Feb 09, 2007 11:50 pm
Location: Sweden

Postby MrPete » Wed Jun 04, 2008 12:15 pm

leo,

Thanks for the advice, the extra gig of ram and a video card aren't too expensive will probably do that soon.

Magweb,

"If five scans are resulting a about 92 MB file and you need 40 scans (do you realy need 40?) the result will be a 736MB final file!!! How do you want to use this giant file?"

I scanned at the higher resolution trying to capture as many details as possible form the model. do I really need 40? I thought so - the aspect ratio of the model forces me into multiple scans along it's length, in this case 5 scans. After that I'm shooting it from 8 different sides - top, bottom, left, right and the four diagonals to eliminate areas that may have been shadowed or tangential to the laser. Perhaps I could get decent results from four sides with spot scanning problem areas. I'm already planning on reducing the resolution to help with file size.

Once completed will take the finished model into Rhino or Pro Engineer and blow it up to make a larger model to use as a pattern for making a physical model. Either program will allow me to slice and dice the model into local sections to aid in model making. Basically take a 19" plastic model and use it as a pattern for making a 4-5 foot long one from fiberglass.
User avatar
MrPete
 
Posts: 31
Joined: Mon Mar 24, 2008 1:17 am

Postby MagWeb » Wed Jun 04, 2008 2:20 pm

Hi again,

If your modell is 19'' (about 480mm) and you used 5 shots per length (means if you used about 2/3 as not overlapping) at 1200x1600 you had 800 horizontal camlines per 100mm. To get an effect of this resolution your laserlinewidth should be less than 0,125mm (100/800) .

Only a reasonable relation of linewidth and resolution improve the results.

Gunter
MagWeb
Moderator
 
Posts: 2499
Joined: Wed Jul 18, 2007 8:48 pm

Postby MrPete » Wed Jun 04, 2008 2:55 pm

magweb,

thanks for doing the math, lets see if it follow it.

19"tall model/5 scans = 3.800 inches per scan

3.8 inch/1200 lines resolution =.0032 inch per line of resolution

So theoretically to get the best possible scan would need a laser line .0032 wide?

That being said I doubt I'll ever be able to produce a beam that fine with the hardware I have. THe laser source is focusable and I've got as fine a beam as possible with it but it's nowhere near .003 wide (probably closer to .060"). will have to measure the beam width, if nothing else will point me to a better match of camera resolution to work with and with such smaller file sizes.

I thought i remembered reading that the software takes the average between the upper and lower edge of the beam to determine location, true?

thanks for the help - I'm learning , slowly but learning
User avatar
MrPete
 
Posts: 31
Joined: Mon Mar 24, 2008 1:17 am

hi

Postby hal » Wed Jun 04, 2008 3:08 pm

Hello MrPete: about your question: "Is there a way of eliminating overlapped points from combining multiple scan objects to minimize file size?", you must wait the next ShapeFusion 2, with its selection and removing triangles tool.

It will come soon :P

Mattia
PC: DELL Alienware 17, Intel I7-4710MQ, 32 Gb Ram, AMD Radeon R9 M290X 4GB.
SO: Win 8.1 64Bit.
CAMs: USB 3.0 CMOS B/N (1280x960, 60 fps).
LENS: 12 mm.
PROJECTOR: DAVID SLS1 ACER K11.
User avatar
hal
Moderator
 
Posts: 1527
Joined: Thu Mar 01, 2007 4:08 pm
Location: Cavaria con Premezzo (VA), Italy.

Postby MagWeb » Wed Jun 04, 2008 4:47 pm

Hi, MrPete

yes, that´s the principle in theory and in ideal - if I've understood DAVID right

(but for you need to have some overlapping areas, you cover the effective 100mm with only about 66% of the cam's vertical means about 800 lines in your case)

The nearer you come to that value the better.

You are right, DAVID calculates the centerline of the beam . But that's a calculated average:
If the lower edge of the beam is at a value of 1 in line 3 and the upper at a value of 9 in line 1, DAVID would calculate the value of 5 to the pixel in line 2.
Now the line moves one pixel. Let´s say it has the lower edge value of 2 and the upper maybe 6 so DAVID calculates 4 to line 3.

................real..............calculated
Line 1...... 9
Line 2...... 6 ..................5
Line 3...... 1 ..................4
Line 4...... 2

The result is some avarage-smoothing similiar to a lower resolution but with a much higher filesize.

If upper and lower edge are on the same pixel, they own the same value. So DAVID calculates "right"

Note: This wasn´t discussed before. So this explanation may have some mistakes. I was inspired by the relation of resolution and objectsize of your scans contrary to the relation I use to scan.

Note to the linewidth:
The line is thinner than you see it (you´re blinded). So lines on darker material seem to be thinner than on white


Gunter
MagWeb
Moderator
 
Posts: 2499
Joined: Wed Jul 18, 2007 8:48 pm

Postby MrPete » Wed Jun 04, 2008 5:02 pm

Note to the linewidth:
The line is thinner than you see it (you´re blinded). So lines on darker material seem to be thinner than on white

Is there a known correlation between visible size and detectable via the camera for the sake of calculating out the resolution to pick to get maximum resolution with minimal file size?

If there isn't a constant or something of that nature is there a way to experimentaly determine the beam width the camera sees?
User avatar
MrPete
 
Posts: 31
Joined: Mon Mar 24, 2008 1:17 am

Postby MagWeb » Wed Jun 04, 2008 5:35 pm

Hi again


Calculating the linewidth by eye is a very subjective thing:

I see my laser being about 1mm. WalterMo has measured the same type of laser LC 650-5-3F at 0,23mm at a distance of 0,5m:

see
http://www.david-laserscanner.com/forum/viewtopic.php?p=1420&highlight=#1420

But I think there could be some tolerances too.

Until now you´ve to get some feeling...

Gunter
MagWeb
Moderator
 
Posts: 2499
Joined: Wed Jul 18, 2007 8:48 pm

Postby MrPete » Fri Jun 06, 2008 2:39 am

Magweb,

Fascinating stuff on how to measure the laser line width, WalterMo's work is outstanding along with his documentation of the process used.


Rather than testing the beam width decided to try testing scan results given the beam I have at different resolutions. From that make a judgement call on results by feature definition and file size.

Look at the edges of the V shaped feature towards the top of the scans and the ribs contained therein for a good comparison of the quality.
Attachments
060508SCANS.jpg
User avatar
MrPete
 
Posts: 31
Joined: Mon Mar 24, 2008 1:17 am

Postby MagWeb » Fri Jun 06, 2008 8:51 am

Hi MrPete

thanks for this comparison.
From res 640x480 to 960x720 your filesize growed 66% and the notable differences are only small.
According to my maths, the ideal linewidth should be 0,23mm in res 640 and 0.15 in res 960 (the part is about 110mm).The real linewidth should be far over that in both cases. (perhaps we need to add a factor to that rule)

It would be interesting to complete the comparison with changed relations at the same res. Means: Scanning the same object in half and total (increased distance object/cam) . There you should see more differences comparing the same objectregion.

Idea to get the linewidth: Why don´t we simply count the pixels our line enlights in camview (by enlarging a shot)? Pixel x mm/pixel = laserlinewidth

Gunter
MagWeb
Moderator
 
Posts: 2499
Joined: Wed Jul 18, 2007 8:48 pm

Postby MrPete » Fri Jun 06, 2008 11:08 am

Magweb,

Funny you mention this it was next on my to try list.

It would be interesting to complete the comparison with changed relations at the same res. Means: Scanning the same object in half and total (increased distance object/cam) . There you should see more differences comparing the same object region.

I believe it was you who had suggested filling the viewable area with the obect to scan to get the best resolution possible. A while back had the object nearly fill the screen and cover the backgound and had isssues with the laser line dropping out, have been assuming there was insufficient length to determine the reference line and as such increased the background size slightly to provide a little more boundary around the object.

Seeing the current counter intuitive results (lower resolution providing better results) need to take a better look at what I'm doing. Perhaps the issue with the laser line not being detected wasn't that I didn't leave enough boundary around the model but the camera settings instead, it was there but he camera as configured just did not see it????

I've been scanning the forums to see if anyone had sugested settings for the camera I'm using - logitech Pro 9000, at the moment it like everything else here is trial and error.

For a laugh thought I'd post a picture of the "scrap" scanner made up primarily of scraps from the workshop and friends junk drawers
Attachments
Slide1.jpg
User avatar
MrPete
 
Posts: 31
Joined: Mon Mar 24, 2008 1:17 am

Postby MagWeb » Fri Jun 06, 2008 1:57 pm

Great if you would do this!

You find example settings for a dark ambient light at:
http://www.david-laserscanner.com/forum/viewtopic.php?t=449

Some off-topic suggestion (but we are very off topic anyhow):
The objects surface is shiny. The effect is visible below the big dot in your testscans. This is caused by reflections. Try to matten the object with powder or reduce the cam's sensitifity.

I believe it was you who had suggested filling the viewable area with the obect to scan to get the best resolution possible.


Yes that´s right,. I´ll be more carefull next time. But you´re the first choosing the extreme: Big with high resolution. More often was the other extreme: Small object in a lower resolution. DAVID simply needs the right balance.

Regarding the quality/resolution/laserlinewidth problem:
The speed of the lasermotion has some influence too:
-To quick means a smeared, broader line.
-Very slow speed (far below one pixel progress per frame) means, the value of the centerline is calculated several times as the line moved very little. The new data have a weight of 70 to 30 of the old. So the recalculation smoothes calculation extremes/errors. This way WalterMo is able to scan such small objects (with a very bad linewidth/objectsize relation).

Gunter
MagWeb
Moderator
 
Posts: 2499
Joined: Wed Jul 18, 2007 8:48 pm

Postby MrPete » Sat Jun 07, 2008 4:52 pm

Magweb,

(yes we're way off topic but it is a good evolution of process implementation)

I started setting up to test different working depths on the camera and quickly realized that my setup is not very friendly in this regard. Going longer it quickly picks up all sorts of exraneous points (threaded rod on the framewok, dark area beyond the background and hole in the floor of the scanner setup) for calibration and end up chasing what it thinks are calibrqation workpoints and playing with camera settings trying to eliminate false target points. Moving the camera in quickly move past the calibration points and require a new reduced scale background to work with and am near the minimum working width for what i'm scanning now.

That being said am going to try experimenting with what is readily adjustable at the moment to improve the scans.

I entered the values you had linked to for the camera settings and was picking up ghost pixellation off the model and tried adjusting the settings, ended up with a brightness setting of 1194 where you had 5020 to eliminate the ghosts but i'm still getting reflection off the model as you had noted. I'm going to spray the model with a white primer my son used on his warcraft figures - very fine mist, doesn't buildup on the surface and is quite flat in finish. After that am going to play with camera settings to get the best effect.
User avatar
MrPete
 
Posts: 31
Joined: Mon Mar 24, 2008 1:17 am

Next

Return to Shape Fusion

Who is online

Users browsing this forum: No registered users and 1 guest

cron