Scanner-Killer Workflow

[Infinite]

Hi all,

I wanted to open a new thread here (hope it's ok) regarding how to use Scanner-Killer created by Helmut Kungl of xyz-rgb. Whether it be on-line or off-line.

The site and service:


http://www.scannerkiller.com/welcome.html (http://www.scannerkiller.com/welcome.html)

Over the last 1-2 years I have gained allot of experience trying to get to grips with Stereo photogrammetry, using Scanner-Killer and I would very much like to share my findings and research to help others to learn the process. As it can be a pain staking journey, as well as quite costly.

The plan is to offer some brief guides on the hardware you would need to buy, how to calibrate the cameras to work with the software and how to process the data generated.

Here's an example of a 3 Camera Set-up. 2 Cameras configured for Active Capture (with a projected noise pattern) and 1 Camera configured for Passive Capture (strobe fire just for colour)

http://www.ir-ltd.net/uploads/cgfeedback/IMG_0842.jpg

Images Captured:

Camera1 (Active)

http://www.ir-ltd.net/uploads/cgfeedback/Camera1_07.jpg

Camera2 (Active)

http://www.ir-ltd.net/uploads/cgfeedback/Camera2_07.jpg

Camera3 (Passive)

http://www.ir-ltd.net/uploads/cgfeedback/Camera3C_07.jpg

Example Disparity Map

http://www.ir-ltd.net/uploads/cgfeedback/Camera1D_07.jpg

Example Capture Slice:

http://www.ir-ltd.net/uploads/cgfeedback/SK-Colour-01.jpg

More to come....

[Infinite]

Starting out and Shopping List.

[Infinite]

Set-up and Calibrations.

[Infinite]

The Captures and Processing the data.

[Kel Solaar]

Posting in a future gold thread :) Thanks for the initiative Lee, looks promising! :]

KS

[rasmusW]

very exiting.. looking forward to see more of your research.

-r

[BrettSinclair]

sponge bob checkered board:p. Great tips Lee. Always amazes me how much you share. Thanks again.

[gjpetch]

Very cool stuff Lee, have you tried projecting coloured noise instead of monochrome noise? I've got the idea that the additional information in coloured noise could result in a better capture, not sure?

[Infinite]

Working with Templates. Extraction and Projection:

[Infinite]

reserved bagsies 2

[Infinite]

Kel: Thanks :thumbsup: Much more to come. Can't keep all this info locked away, it seems a waste, plus I don't want others to go through the rigmarole I did in learning all this stuff.

rasmusW: :)

Brett: Thanks, hope you find it useful. I plan on getting in a space suit and some other costumes soon! gonna capture them! not a proper Nasa suit but sommat like this:

http://www.ir-ltd.net/uploads/cgfeedback/729.jpg

gjpetch: No I haven't tried it in the studio but I have a virtual set-up were I can render out fake set-ups and process off-line. Using virtual camera and information in Max or Lightwave. I will share some scenes soon. Problem with this software is it can get confused with what is background and foreground, it's a constant pain, so sometimes things like legs, arms or hands (spindly things) get chopped off or confused.

I will give it a go! Good idea thanks.

[PascalR]

This great thread is now sticky and shall soon be plugged, thanks Lee for sharing your knowledge on the subject, very interesting stuff!
:thumbsup:

[PascalR]

:sw:plugged:sw:

[Infinite]

Thanks Pascal! Much more info to come :)

[meshmasters]

The Captures and Processing the data.



One of the draw backs with this type of technology, is the software failing to recognize what data is background data and what data is foreground data. Therefore it can get confused, and you can loose parts of the capture. This happens on overlapping body parts, arms being cut off or hands.

I am still working on trying to fix this, but it is not easy and somewhat a case of pot luck.




Lee,

Just a thought regarding this issue.

Why not move the subject to be captured further away from the background and at the same time, light the background so that it is hotter than the rest of the scene (if that makes sense) I would light the blue screen in the same way as I would if I were doing a blue/green screen for film/tv -- in other words, light it so that it is bright and even. Doing this, should diminish the pattern being projected onto the background and therefore make it less likely that it will be picked up and processed. Mind you, I don't know if this will work or not... but it is worth a try, why not give it a go?

If you try this, let me know how you get on, I'd be interested.

Cheers,

Joe

[Infinite]

Lee,

Just a thought regarding this issue.

Why not move the subject to be captured further away from the background and at the same time, light the background so that it is hotter than the rest of the scene (if that makes sense) I would light the blue screen in the same way as I would if I were doing a blue/green screen for film/tv -- in other words, light it so that it is bright and even. Doing this, should diminish the pattern being projected onto the background and therefore make it less likely that it will be picked up and processed. Mind you, I don't know if this will work or not... but it is worth a try, why not give it a go?

If you try this, let me know how you get on, I'd be interested.

Cheers,

Joe

Hi Joe, sadly no. This doesn't do much. It also doesn't solve issues when arms are in front of a body. I have run various virtual tests using the software trying everything imaginable. Also doing the same in the studio. It seems to be a constant issue.

With the on-line service for SK and the off-line software application you have the option to mask the background, effectively doing what you suggest above but manually. This also has varied results, sometimes it helps, sometimes it doesn't there seems to be no real solution. Kind of like a lotto.

Sometimes I get perfect captures, other times various limbs are missing, even after masking and cranking up the processing iterations :(

But I think it should be enough to get by with, especially if you get multiple captures of the same person, say on a turntable.

The annoying thing is, to a human eye it's blatantly obvious what is foreground, midground and what is background, so this is a very frustrating issue.

For example this was a pretty good 360 set, only on one of the captures did the arm break:

http://www.ir-ltd.net/uploads/cgfeedback/SK-Grabs2-01.jpg

On a naked model it's more pot luck.

[meshmasters]

Hi Joe, sadly no. This doesn't do much. It also doesn't solve issues when arms are in front of a body. I have run various virtual tests using the software trying everything imaginable. Also doing the same in the studio. It seems to be a constant issue.

With the on-line service for SK and the off-line software application you have the option to mask the background, effectively doing what you suggest above but manually. This also has varied results, sometimes it helps, sometimes it doesn't there seems to be no real solution. Kind of like a lotto.

Sometimes I get perfect captures, other times various limbs are missing, even after masking and cranking up the processing iterations :(

But I think it should be enough to get by with, especially if you get multiple captures of the same person, say on a turntable.

On a naked model it's more pot luck.


Ahh, I see.... bugger! :-/

Well, it was worth a try :-)

Cheers,

Joe

[vargatom]

This stereoscopy based scanning stuff is very interesting... but as it seems we're still not at the point where it would be a simple and clear process. We've been thinking about scanning, if only for general reference for our sculptors and blendshape modeling, but despite all your advice and insight, this appears to be something not yet worth investigating, at least until we have a longer period without funny deadlines.

Also, XYZRGB uses laser scanners as well, right? But I suppose such a device that can do a full body scan or get details at the millimeter level is still very, very expensive. I don't think there are any of those in Europe yet, and the smaller ones I've seen available commercially are all far too limited in their accuracy.


Also, at FMX there was a talk on creating Clu from the new Tron movie by Matthias Wittmann. He said they've scanned Jeff Bridges using two methods, because of the various issues associated with the specific methods.
First they took a lifecast of his head and digitized the result, but as the process took 30 minutes, he inevitably loosened his jaw and thus the result was distorted, they couldn't use it on its own to deduce the position of the teeth.
Then they used a Lightstage from Paul Debevec, but this has another problem, the lights are very bright and the subjects inevitably try to tighten their eyes, distorting the face from the neutral pose once again. It's also far from being as accurate as the lifecast which has pore level details.
So they combined the two methods to build a single model of the 60-years old Bridges, and then used it as the base for the younger version for Clu.

[Infinite]

This stereoscopy based scanning stuff is very interesting... but as it seems we're still not at the point where it would be a simple and clear process. We've been thinking about scanning, if only for general reference for our sculptors and blendshape modeling, but despite all your advice and insight, this appears to be something not yet worth investigating, at least until we have a longer period without funny deadlines.

Also, XYZRGB uses laser scanners as well, right? But I suppose such a device that can do a full body scan or get details at the millimeter level is still very, very expensive. I don't think there are any of those in Europe yet, and the smaller ones I've seen available commercially are all far too limited in their accuracy.


Also, at FMX there was a talk on creating Clu from the new Tron movie by Matthias Wittmann. He said they've scanned Jeff Bridges using two methods, because of the various issues associated with the specific methods.
First they took a lifecast of his head and digitized the result, but as the process took 30 minutes, he inevitably loosened his jaw and thus the result was distorted, they couldn't use it on its own to deduce the position of the teeth.
Then they used a Lightstage from Paul Debevec, but this has another problem, the lights are very bright and the subjects inevitably try to tighten their eyes, distorting the face from the neutral pose once again. It's also far from being as accurate as the lifecast which has pore level details.
So they combined the two methods to build a single model of the 60-years old Bridges, and then used it as the base for the younger version for Clu.

No it's really not an easy process. There's quite a steep learning curve and it takes time to get to grips with. You need patience and resources.

This type of capture is ideal for reference and for sculptors to use. Ideal for FACS capture and for studying Human anatomy, in 3D instead of 2D even if it's just slices for separate angles to study, or re-project from.

XYZRGB have laser scanners yeah but I have yet to see anything that can do a full body using that method that can produce highly detailed captures. They did allot of laser scans of plaster casts like you mentioned.

Most studios that have the money, like Gentle Giant still use those ancient Cyberware scanners. They swear by them but the results are pretty low-res. The beauty of them is speed I think. All well and good if you have a few hundred thousand dollars to spare and LOTS of room.

Interesting what you say about Clu, thanks for sharing that. From what I have heard the Lightstage data is far from perfect and needs allot of re-work. One of the reasons Aguru went bust as the data supplied isn't that easy to use. In the right hands though I would imagine it's ideal. To hire the Lightstage is super expensive but the results are very detailed, I also heard that about the lights being bright. If they can, they get their subjects to wear tinted contact lenses to cope with the brightness.

I've also heard rumours that the Lightstage 6 (full body capture) is defunct and gathering dust, such a shame as it looked super impressive.

Lifecasts and laser scans of cast is the best way to go for more accurate details I think.

[Intervain]

very interesting thread Lee

[mala]

Hey Lee,
Very interesting thread, thanks for sharing this info.
All a bit out of my price range but some of your tips here can be applied to improving my experiments with Photoscan.

I take it that using the 2.5mm jack splitters bodge works fine then,and your 2 cameras fire at exactly the same time, saving you the cost of another pocket wizard?

Interesting to hear about the process for Clu...one thing i still don't really get with the light stage is why it has to be SO bright...of course it has to be bright enough to shoot at fast speeds and cleanly retrieve detail...but surely some balance can be achieved between this and 'blinding' the subject?

Don't think i have any ideas on the problem of separating subject from background with your full body scans other than what Joe 'meshmasters' has said..a bright,evenly lit backdrop..

looking forward to reading more,
Cheers.

[Infinite]

Thanks Magdalena : )

Hey Merry, I hope some of this stuff might help with Photoscan. I would be interested to see. I'm not sure if the shutters can be mechanically synced perfectly (I think that is impossible) but using the Flash sync technique you can do it through the Cameras sensors. That's the trick and could improve Photoscan captures a great deal, maybe...

Yeah I don' quite get why the Lightstage needs to be so bright. It shouldn't need to be. Disney don't do that, and they use the same Flash sync technique as above with just as a good a result in details (with no normal map capture though) but they have a super powerful "Mesoscopic High Pass Filter" technique. It beats the crap out of Photoshop's High Pass filter. Nothing is publicly available though and these guys aren't very open about what they do sadly.

The background separation isn't an issue with SK. It's just useful to be able to quickly mask the background from subject, actually a simple Black Backdrop is more effective. I was just trying to find a way to use their new built in Mask detection feature. It still wont solve the problem with the algorithm getting confused when body parts over lap etc.

Cheers,
Lee

[madmatt]

Thanks Lee for sharing all your testing Work, it's really interesting for me!!!

Bye

[vargatom]

Pictures on the FMX facebook page are online, there's one from Bridges' head:

https://fbcdn-sphotos-a.akamaihd.net/hphotos-ak-ash4/224354_10150249202901745_63281456744_9056059_11997 47_n.jpg

[Cory]

Crazy amount of detail in that scan. Would DD extract the displacements/bump maps from this head for the in movie head? I guess then after that they would need to change them to suit a younger version of bridges?

It's always cool seeing these images from such a well made movie, I want more. :drool:

[Infinite]

No problem Madmatt.

Pictures on the FMX facebook page are online, there's one from Bridges' head:

https://fbcdn-sphotos-a.akamaihd.net/hphotos-ak-ash4/224354_10150249202901745_63281456744_9056059_11997 47_n.jpg

Thanks Tamas. Do you have a link to that Facebook page?

The details are great! DD must have allot of tools at their disposal. All the more surprising regarding the final results in Tron.

[Infinite]

Some might find this interesting, related to those types of Lightstage 3 Camera Position captures http://www.cmlab.csie.ntu.edu.tw/~liubiti/Merge/index.html

[companioncube]

Thanks Tamas. Do you have a link to that Facebook page?

http://en-gb.facebook.com/photos.php?id=63281456744

some kind of other scan from FMX, not sure from what talk
http://a6.sphotos.ak.fbcdn.net/hphotos-ak-snc6/228396_10150248404876745_63281456744_9050858_42984 80_n.jpg

[Infinite]

http://en-gb.facebook.com/photos.php?id=63281456744

some kind of other scan from FMX, not sure from what talk
http://a6.sphotos.ak.fbcdn.net/hphotos-ak-snc6/228396_10150248404876745_63281456744_9050858_42984 80_n.jpg

Cool! looks like Paul Debevec talking about their 4D capture process?

[vargatom]

That scan from Tron was actually the less detailed one, from the lightstage capture. The image isn't big enough, I sat quite close and it was more obvious from there. I know Debevec has some method to extract normal maps from perhaps the two specular only passes created with polarized light or so, but it's not the same as having a laser scan with sub-millimeter accuracy.
Also, the markers on his face are there to compensate for any distortions. They've measured the distances between the markers and then they've tweaked the mesh to conform to these measurements.

They've used the lifecast's scan for details reference, but as far as I know the young head was completely manual work, sculpted in Mudbox and going through a lot of revisions.
They've however referenced the pore patterns and wrinkle placement from the old head, and it was also important for their facial capture solver which needed the face to analyze marker positions in the image data.

Yeah and I haven't linked to the facebook page because there weren't any other photos that I've found interesting or related to this topic ;)

[madmatt]

Cool! looks like Paul Debevec talking about their 4D capture process?

He looks like Paul I'm pretty sure!!!

Bye

[El Burritoh]

Lee sure knows how to start a thread of Pure Awesomeness. I've got to read back over this more carefully...

[Olivier Renouard]

Nice Ian Solo scans!

Seriously, very interesting thread, will come back regularly to check.

[NextDesign]

Cool! looks like Paul Debevec talking about their 4D capture process?
That is indeed Paul Debevec presenting his paper "Temporal Upsampling of Performance Geometry using Photometric Alignment"

See more here: http://gl.ict.usc.edu/Research/TempUpsampling/

[Infinite]

That is indeed Paul Debevec presenting his paper "Temporal Upsampling of Performance Geometry using Photometric Alignment"

See more here: http://gl.ict.usc.edu/Research/TempUpsampling/

Thanks John, yeah seen that. Very cool, I presumed the one AJ posted was something newer at FMX? Was hoping to try and find some new goodies to look at on-line. As their site hasn't been updated in a while.

[morbo]

Hey Lee,

Thank you for helping demystify sk :)

I've been trying to wrap my head around it for a couple of months, and I'm glad to see that I'm not the only one who finds it to be a painstaking process.

I'm envious of your 3 camera setup :) Is that the only way you can do a live extraction with colour?

I had a demo by Helmut, and I can't remember if he had a projector on me or not...but I do remember he only had 2 cameras...and the scan looked decent.

I haven't been able to match it myself.

As you mentioned it can get expensive, and I've bought almost all of the gear except the projector. ...and instead of pocket wizards...I used Phottix Atlas triggers (partly because they were cheaper, and I couldn't find the proper Pocket wizard cables for my Canon 60d's anywhere)

Does the projector make that much of a difference? (I ask only because thats another $1000 in this bankaccountkiller project :) )

[Infinite]

Hey Lee,

Thank you for helping demystify sk :)

I've been trying to wrap my head around it for a couple of months, and I'm glad to see that I'm not the only one who finds it to be a painstaking process.

I'm envious of your 3 camera setup :) Is that the only way you can do a live extraction with colour?

I had a demo by Helmut, and I can't remember if he had a projector on me or not...but I do remember he only had 2 cameras...and the scan looked decent.

I haven't been able to match it myself.

As you mentioned it can get expensive, and I've bought almost all of the gear except the projector. ...and instead of pocket wizards...I used Phottix Atlas triggers (partly because they were cheaper, and I couldn't find the proper Pocket wizard cables for my Canon 60d's anywhere)

Does the projector make that much of a difference? (I ask only because thats another $1000 in this bankaccountkiller project :) )

Hi Harry,

No worries : )

Yeah the 3 Camera system is only really useful for far away full body or mid torso shots. As long as you can synchronize your flash to your 2 camera setup then it should be fine for faces, even torso shots. It's all about pattern and high contrast, so the type of lenses you use are also really important, and these tend to be far more expensive than the cameras. So you can imagine the more Cameras you have the more lenses you need. It gets very expensive.

This same methodology applies to Agisoft and Photofly.

I really recommend those pocket wizards and mulit-max. You can't synch your Cameras mechanically but you can do it by the strobe/flash! :thumbsup:

It's really important your cameras are synced for the extraction process. As any motion blur can hinder the process.

I really recommend the Pocket Wizards over anything else, they are rock solid realibale. After taken 100's of images I haven had any fail or not sync.

For cables you can buy a converted "Headphone Headset Adapter 3.5mm to 2.5mm Y-Splitter"

Projectors aren't cheap but can be useful and you can always use them for a home cinema when they aren't busy :beerchug:

Soon I am hoping to ditch the projector for some kinects :D and convert 2 of the cameras to Infrared. Which I know is possible from the virtual tests I have made.

Flash Simulated

http://www.ir-ltd.net/uploads/cgfeedback/sml/SK-plain_Proces.jpg (http://www.ir-ltd.net/uploads/cgfeedback/SK-plain_Proces.jpg)

Projector Simulated

http://www.ir-ltd.net/uploads/cgfeedback/sml/SK-sim-projected-pattern.jpg (http://www.ir-ltd.net/uploads/cgfeedback/SK-sim-projected-pattern.jpg)

IR Simulated

http://www.ir-ltd.net/uploads/cgfeedback/sml/SK-kinect-IR-_Proces.jpg (http://www.ir-ltd.net/uploads/cgfeedback/SK-kinect-IR-_Proces.jpg)

Benefits of IR capture over standard projector:

No interference from the noise pattern in the colour capture
Invisible pattern
No bright projection into subjects eyes
Dirt cheap
IR Laser Pattern always in focus <- HUGE benefit
You can hook up 4-5 Kinects using a few PC's (to cover various angles) for even less than a standard projector price!
You can use a certain type of dark cloth that wont pick up the IR pattern, even better for masking and extraction!
Fun!


Draw back of Standard Projector:


Heavy and cumbersome
Noisy
Get very Hot
No external control over image display, except via PC, so pattern always on
Expensive
Pattern interferes with Colour Capture
Can blur and defocus in different capture quadrants and depth
Restricted resolution - price


This is really interesting and related to IR with Kinects - http://www.futurepicture.org/?p=97

So for full body capture, yes an Active Pattern makes a big difference. It all depends how good your lenses and cameras are, the more pattern and contrast you have the better the results will be. For far away shots, noise really helps.

Best,
Lee

[morbo]

Nice!

Thanks for the thorough reply ;)

I've changed up my setup a fair bit because of your postings...thank you!

Nice idea about using the splitter for your pocket wizards ...I can't believe I didnt think about it...I ended up being two Atlases instead...but the splitter provides a better sync.

Now I have an extra one for another stereo pair (once I've figured out how to work my first pair)

The infrared setup sounds neat! Please keep us informed...and if you need extra testing...let me know.

I have a quick question about the calibration of the cameras....

Using the checkerboards to calibrate is the procedure,
Does the lighting have to stay consistent between the checkerboards and the actual model/subject?

[Infinite]

Nice!

Thanks for the thorough reply ;)

I've changed up my setup a fair bit because of your postings...thank you!

Nice idea about using the splitter for your pocket wizards ...I can't believe I didnt think about it...I ended up being two Atlases instead...but the splitter provides a better sync.

Now I have an extra one for another stereo pair (once I've figured out how to work my first pair)

The infrared setup sounds neat! Please keep us informed...and if you need extra testing...let me know.

I have a quick question about the calibration of the cameras....

Using the checkerboards to calibrate is the procedure,
Does the lighting have to stay consistent between the checkerboards and the actual model/subject?

Hi Harry, Glad some of the info was useful : )

Yes it is fairly important to keep consistent lighting between shots but just for Calibration. Also important to try and remove any sheen, so a matt finish is useful on the boards.

The lighting on your subject after calibration can be anything, as long as it works well when extracting the information. i.e. Projector pattern, flash, IR etc

You can even do Normal Map acquisition, if you have a Global Illumination lighting rig, like Lightstage 6.

You would need 5 very rapid captures:

Active (projector, IR shot)

GI Lighting shot (your main texture) - All Lights
Left Light group
Top Light group
Front Light group

You can then do image division on the GI captured image using the other lighting shots to generate synthetic right, bottom and back lighting shots, then slot them all into a blank PSD in the RGB channels to get XYZ=RGB :sw:

You can then use this data to bump, emboss modify the mesh further for more high frequency details that are stored in the RGB channels.

Not an easy process but possible.

Best,
Lee

[Infinite]

My first ever Water Tight 360, Full Colour, Full Body Capture! A bit rough around the edges but the pipeline is pretty sound. Learnt allot during this process. Not bad for just 8 pairs of processed images.

Scanner-Killer, 360 Full Colour Capture.

3 Camera Setup (2 Active, 1 Colour) 8 Captures. 10,000 Polygon Subdivision Model with UV Map, 6Kx6K Texture

Warning Nudity!

http://www.ir-ltd.net/uploads/cgfeedback/sml/SK-Pose06-Compile-02.jpg (http://www.ir-ltd.net/uploads/cgfeedback/SK-Pose06-Compile-02.jpg)

I need to correct the face, sadly I didn't mix the 3/4 images well enough, so looks a bit planar mapped. Easy to fix.

[madmatt]

Only 8 photos!!!? Great results...

Bye

[Infinite]

Only 8 photos!!!? Great results...

Bye

Well each capture is processed from a set of 3 Camera images, X8 different angles so it's more like 24 images but 8 of those are just used for colour the other 16 are for the stereo data. But it's not like Photofly or Agisoft where you have to move all around the subject and take dozens and dozens of images, plus I think the final results are allot cleaner and sharper, you have more control as well.

[madmatt]

Well each capture is processed from a set of 3 Camera images, X8 different angles so it's more like 24 images but 8 of those are just used for colour the other 16 are for the stereo data. But it's not like Photofly or Agisoft where you have to move all around the subject and take dozens and dozens of images, plus I think the final results are allot cleaner and sharper, you have more control as well.

Definitely!!! Thanx for share results & tests...

Have a nice weekend...

[PascalR]

great work Lee, very nice result indeed

[Infinite]

great work Lee, very nice result indeed

Thanks Pascal, still early days yet. Going to be able to test the Infra Red method this week. Will share soon.

Update:

Just fired up the Kinect (after a painful install I might add) along with an old converted 400D with an IR Filter 720NM. It picks up the IR pattern perfectly, and I am working with some developers to find a way to fire up multiple kinects (more than 2) to spread the laser pattern and to make it more dense ready for Active capture and processing by Camera 1 and 2.

The purpose for these tests is to allow Camera 3 to be Active Pattern free.

[morbo]

Nice Work on your full body scan there!

Thanks for your reply Lee,

The reason why I was asking about the lighting wasn't so much for aesthetics...though, all your points are pretty awesome and I will try them out.

...I was thinking, that since the checkerboard patterns are used to define lens/camera properties for sk,

I was hoping to see if it was possible to calibrate my cameras, which both use prime lenses (50mm 1.8), then without changing the setups at all...move the rig and or just the sidearm with the cameras (sans tripod) around as opposed to bringing the subject to my rig.

I would think that, nothing would change...since lenses are fixed, focus rings are taped down, tripod heads are locked.


Thoughts?

[Infinite]

Nice Work on your full body scan there!

Thanks for your reply Lee,

The reason why I was asking about the lighting wasn't so much for aesthetics...though, all your points are pretty awesome and I will try them out.

...I was thinking, that since the checkerboard patterns are used to define lens/camera properties for sk,

I was hoping to see if it was possible to calibrate my cameras, which both use prime lenses (50mm 1.8), then without changing the setups at all...move the rig and or just the sidearm with the cameras (sans tripod) around as opposed to bringing the subject to my rig.

I would think that, nothing would change...since lenses are fixed, focus rings are taped down, tripod heads are locked.


Thoughts?

Yes that should work ok, as long as your Cameras don't move position relative to each other.

I tested on a mobile Fig Rig which worked a treat, great for super high res shots:

. .
-

http://www.ir-ltd.net/uploads/cgfeedback/sml/SK-Mobile-01a.jpg (http://www.ir-ltd.net/uploads/cgfeedback/SK-Mobile-01a.jpg)

[morbo]

Of course you've already tested a mobile rig ;)

Nice!
That's great to know that it's possible!

[morbo]

Hey Lee,
so I just got myself a projector...and did a quick test...looks great so far...

BUT...I have a question about the calibration of the checkerboards...

Do you adjust the focus on the cameras when you take pictures of the checkerboards in depth? meaning the close ups...and the distant ones?

Also, with the standalone version of sk...I know you can use 3 cameras...
how do you get the third to take a shot without the noise pattern being projected? Thanks

H

[Infinite]

Hey Lee,
so I just got myself a projector...and did a quick test...looks great so far...

BUT...I have a question about the calibration of the checkerboards...

Do you adjust the focus on the cameras when you take pictures of the checkerboards in depth? meaning the close ups...and the distant ones?

Also, with the standalone version of sk...I know you can use 3 cameras...
how do you get the third to take a shot without the noise pattern being projected? Thanks

H

Hi Harry,

When you first Calibrate you only focus once, don't change the focus. For the main middle position of the board you focus, then for the next near set you move the board forward a little from the main focus, then for the back set you move back from the main focus point again. Keep the focus set.

Do you have the stand alone version? I can put you in touch with Helmut Kungl if you are interested in a copy? It's not cheap though.

With a 3 Camera setup, you can calibrate 3 cameras together. 1+2 take Active Noise capture, 3 takes colour. With the stand alone software if it reads 3 Camera folders it can pick up a 3 camera setup and process accordingly, using the 3rd camera set of images instead of 2nd camera set to project the textures.

I hope that helps.

Cheers,
Lee

[morbo]

haha...I'd love to get the standalone...but yes it is pretty pricey...I've spoken to Helmut about it. (out of my range for now :) )

The reason why I was asking was because I was thinking about using a third camera on my rig.
-have it trigger at the same time, and then morph (via Mari or whatever) the texture to match the speckled projector texture.

Before I went ahead and borrowed/bought another camera (eeek!), I wanted to figure out the logistics of it all.

How do you make it so the third camera doesn't pick up the speckled projector textures? since you cant put a trigger on the projector (can you?)

- or do you shoot the third camera at a shutter setting too high for the projection?

- or...does your strobe fill in the speckles?

none of them sound like they'd work

(am I missing something totally obvious?:dance:)

Thanks Lee for all your help!
MUCH APPRECIATED!!!!clap

[Infinite]

haha...I'd love to get the standalone...but yes it is pretty pricey...I've spoken to Helmut about it. (out of my range for now :) )

The reason why I was asking was because I was thinking about using a third camera on my rig.
-have it trigger at the same time, and then morph (via Mari or whatever) the texture to match the speckled projector texture.

Before I went ahead and borrowed/bought another camera (eeek!), I wanted to figure out the logistics of it all.

How do you make it so the third camera doesn't pick up the speckled projector textures? since you cant put a trigger on the projector (can you?)

- or do you shoot the third camera at a shutter setting too high for the projection?

- or...does your strobe fill in the speckles?

none of them sound like they'd work

(am I missing something totally obvious?:dance:)

Thanks Lee for all your help!
MUCH APPRECIATED!!!!clap

Hi Harry, I explain about the 3 Camera process at the start of the thread :)

In theory the Strobes should wash out the pattern (although not easy to do) 1+2 use the same exposure settings, 3 uses something else and you use a Multimax to delay the Strobe trigger.

Best,
Lee

[Infinite]

Not much of a major update but got a new piece of code today to be able to switch on Multiple Kinects (not capture with them, just to switch on 5 or more, to get access to the IR laser output)

IR captures work very well in simulation, so only time will tell if I am "completely bonkers, or not" on this idea.

Plus new Cameras and Lenses on the way. Now up to 18MP capture with 21MP colour processing.

2 Overlapping Patterns. 3 more Kinects on the way:

http://www.ir-ltd.net/uploads/cgfeedback/IRIMG_4970.jpg

Oh and for fun (Noooo.. it's not Christmas time!):

http://www.ir-ltd.net/uploads/cgfeedback/IRIMG_4966.JPG

[piek]

You must have more kit than Jessops now!

In the last edition of US Wired there was an article on Kinect. Supposedly, the next software upgrade will have new facial motion capture.

[moonjam]

Haha! That infared photo is very cool, I'd love to have a go at that! :D

[BrettSinclair]

Blimey that looks nippy out. Kinda weather where you can cut glass with your nips:p

[Infinite]

You must have more kit than Jessops now!

In the last edition of US Wired there was an article on Kinect. Supposedly, the next software upgrade will have new facial motion capture.

Interesting, I can't wait until the newer version comes out (when ever that might be?) in HD it could be stunning! Currently for scanning (with JUST the Kinect) the results are just too low res and noisy.

moonjam: It's great fun. Like looking at the world through new eyes!

Brett: HaHa! yeah. It also looks like candy land!

[Infinite]

Messing about with 4D Capture. This was the results of the 1st Calibration and 1st Capture. Never did this before but it was so easy to set-up. Had to re-time the HD footage by a few frames to sync.

http://www.youtube.com/watch?v=TjbHT9Yh468

HD Colour (IR Kinect hack) version to come..

[piek]

For a bit of 'messing about', that is seriously good. The amount of reference info you get from these streams is immense. So, is mesh tracking and custom geo binding the final step of your self sufficiency?

[Infinite]

For a bit of 'messing about', that is seriously good. The amount of reference info you get from these streams is immense. So, is mesh tracking and custom geo binding the final step of your self sufficiency?

Hi Pete, I tend to exaggerate when I use terms like 'messing about'

The set-up, calibration and capture took about 10 minutes.

Yeah the next stage is tracking with the colour information and then warping another low-res generic mesh using ICE (possibly) ICE might not be capable of too much in this regard, as you really need some kind of optical flow algorithm or something like Dimensional Imaging have written (for their DI4D) themselves, which is REALLY powerful.

[gjpetch]

Kick *** stuff Lee! Soon movie studios won't have to go to Weta for performance capture, they can just call you up!
Infrared photography messes with my head, I once pulled apart an old webcam and found it had an IR filter that could be removed easily; so weird seeing all this light that we're blind to normally. I've also heard that after eye surgery some people report being able to see into the infrared spectrum..... anyway, I'm getting a bit off topic. Awesome as always Lee!

[Infinite]

Kick *** stuff Lee! Soon movie studios won't have to go to Weta for performance capture, they can just call you up!
Infrared photography messes with my head, I once pulled apart an old webcam and found it had an IR filter that could be removed easily; so weird seeing all this light that we're blind to normally. I've also heard that after eye surgery some people report being able to see into the infrared spectrum..... anyway, I'm getting a bit off topic. Awesome as always Lee!

Thanks Greg, yeah IR and UV stuff is pretty weird lol

[Infinite]

Some new Infra-red 4D Capture tests:

http://www.youtube.com/watch?v=LZm73zi2UKY

http://www.youtube.com/watch?v=ePv9lHdpumA

Very early days, lots of issues to overcome but the beauty is being able to capture refined geometry (not show here) along with globally illuminated colour streams using a 3rd Camera.

Had issues with the focusing on the cameras in poor light. Have to work on that and yes the data is very noisy because of this.

Shot at 1080p at 25fps. The frame rate is undesirable. I will try some tests using 720p at 60fps. They can yield slightly better motion, reducing noise and MB.

used 2xCanon 550D's and 3x Kinects, pitch black room!

[piek]

You should be on commission, because I'm sorely tempted to inquire about a license.

The colour info that you're grabbing, is it distinct/strong enough to do mesh tracking?

[Infinite]

You should be on commission, because I'm sorely tempted to inquire about a license.

The colour info that you're grabbing, is it distinct/strong enough to do mesh tracking?

lol ! :thumbsup: It's worth it.

It would just be good to get others on board! It's tricky doing this stuff on your own as there aren't many others you can turn to for advice or tips. When I Google I find myself these days!

However the customer support and training I have got from Helmut Kungl has been the best I have ever experienced in the CG industry (literally 100's of email correspondences)

That's why I bang on about the software so much. It's fantastic but you need time and patience to get experience using it.

I'm still running more tests at the moment, I'm waiting to find out about some LED Strip lights, if these give off as little heat, IR and UV as they say they do, then they should be ideal to illuminate your content without lighting the IR captures, just the colour. I'm also experimenting with IR Projection from a standard old school projector and 850nm IR Filter, using a pattern Laser Printed onto OHP film. Waiting to see how that goes.

Problem with the Kinects is due to the diffraction through the Laser etched pattern that is used to create the Kinects structured pattern, there is allot of noise and fizzing from the laser, which is undesirable.

It is possible to just do straight Passive capture using 2x standard DSLRS, you just need to get good lighting and great focus. The results can be a little noisy though.

[Infinite]

180 ear to ear 4D Capture, using 4x Canon 550D's at 1080p.

Very scrappy, no smoothing or post processing but looks kind of promising. Only down side is this is using an Active noise pattern which washes out any chance of colour. Still working on IR projection.

Other issue ...this is 5GB's of streamed data! No CUDA power behind this. Streamed using Thiago Costa's OBJ Sequence Importer.

There is a bad seam due to rushed Calibration plus no noise or artefact cleanup.

http://www.youtube.com/watch?v=42P_Vp05DAw

[hsmith1960]

Lee,

For the past week, I've been following the scanner-killer workflow conversations, and I've never seen such creativity! Wow!! Thank you for doing a lot of the hard work to make this process work (hopefully) for newcomers like myself.

I'm on a pretty tight budget, so I was quite interested in your recent idea (August 4, 2011) to use an overhead projector and laser printed OHP film, to provide your noise pattern. I'm interested in sculpture, and I just need to capture the topology from 360 degree scans of live human subjects (the data for UV maps would probably be superfluous)

I'm thinking of using an old 4"x5" photo enlarger (available pretty cheaply on ebay) and mounting it for horizonal projection. The OHP film would go in the negative holder of the enlarger. The only light for the subject would come from the enlarger.

Here's the wild element of my plan: replacing the photo enlarger lamp with a high intensity LED that can be triggered for a short interval to provide the noise pattern. Regarding the interval, maybe less than 1/4 second, if I can get a LED bright enough. I've seen 30W lamps rated at 2100 lumens.

If there's not much ambient light, this might eliminate the need to synchronize the cameras closely, but I'll still need some scheme to do that (a small detail that probably will end up taking a lot of my time...)

I'm thinking the heat generated from the LED could be lessened because of the short duty cycle.

Before I go any farther, do you have any thoughts?
Thanks in advance for any help.

Harry

[Infinite]

Lee,

For the past week, I've been following the scanner-killer workflow conversations, and I've never seen such creativity! Wow!! Thank you for doing a lot of the hard work to make this process work (hopefully) for newcomers like myself.

I'm on a pretty tight budget, so I was quite interested in your recent idea (August 4, 2011) to use an overhead projector and laser printed OHP film, to provide your noise pattern. I'm interested in sculpture, and I just need to capture the topology from 360 degree scans of live human subjects (the data for UV maps would probably be superfluous)

I'm thinking of using an old 4"x5" photo enlarger (available pretty cheaply on ebay) and mounting it for horizonal projection. The OHP film would go in the negative holder of the enlarger. The only light for the subject would come from the enlarger.

Here's the wild element of my plan: replacing the photo enlarger lamp with a high intensity LED that can be triggered for a short interval to provide the noise pattern. Regarding the interval, maybe less than 1/4 second, if I can get a LED bright enough. I've seen 30W lamps rated at 2100 lumens.

If there's not much ambient light, this might eliminate the need to synchronize the cameras closely, but I'll still need some scheme to do that (a small detail that probably will end up taking a lot of my time...)

I'm thinking the heat generated from the LED could be lessened because of the short duty cycle.

Before I go any farther, do you have any thoughts?
Thanks in advance for any help.

Harry

Hi Harry, thanks for posting and for your kind comments. Your idea sounds very good. I would be interested to know which kind of 4"x5" photo enlarger you might be going for and if they currently produce modern versions? Also what would be the high intensity LED lamp you would be going for? triggering that sounds like a good idea.

If your subjects are static, syncing shouldn't be too much of an issue. But don't forget you can always flash sync with the projector (if you can use a controllable LED), with your cameras set at 1/4 speed. You would need a dark environment though. But a noise pattern can sometimes work OK in an ambient lit environment.

The most important thing is high contrast and good crisp images. Lenses are very important.

The issue I have faced with my IR capture is, most DLP, LCD or LED projectors despite there 2000-3000 ANSI Lumins output just don't generate enough IR because of the type of bulbs they use. So they are useless with an IR filter or cold mirror as too little IR light gets through. I also tried with a slide projector and school projector but both had issues. I haven't tested them fully for standard noise projection though. They should be good at that.

And the Kinects just aren't designed for Scanning, the pattern looks like it was designed purely for tracking. It is inherently noisy because of all the little dots. They need to be in a more square, uniform pattern.

The only true IR projector on the market is something called the DLP Lightcommander but is has some weaknesses. Low Lumen output, XGA etc

I'm super interested to see how you get on, please do share your findings here. I hope this thread has helped in some way.

Best,
Lee

[hsmith1960]

Hi Lee,

Thank you for your encouragement. As the saying goes, when you're inside your own head, you're behind enemy lines. Discussing issues with folks who share a common interest is a lot more efficient!

Regarding your questions: here's a ebay item (I hope it's still up when you check it out, the bidding ends soon) for an old 4x5 enlarger:

http://cgi.ebay.com/Beseler-4x5-photo-darkroom-enlarger-extras-/380359955745?pt=LH_DefaultDomain_0&hash=item588f385521

Although the analog market for photos has shrunk dramatically, there are still large format enlargers in production. Here's a link to a discussion that lists most of the decent manufacturers and prices:

http://www.largeformatphotography.info/forum/showthread.php?t=3557

Here's a link to a larger LED that I mentioned (100 W vs. 30 W):

http://www.ledssuperbright.com/30w-led-c-18/100w-white-high-power-led-p-226

Regarding your work with projecting IR noise, what about using a high power IR LED in a photo enlarger? I imagine you'd have to change the focus a bit (because the lens wasn't designed for that short a wavelength), but in theory, you could get the sharpness and the contrast you need, as well as enough radiant energy (I think that's what they use to measure "brightness" in IR)

Here's a link to a IR LED manuracturer: http://www.ledtronics.com/Products/ProductsDetails.aspx?WP=1737

On another note, how did you get those 12 shots of your first 360degree model without a lot of movement from the model? Was she on a rotating platform?

At any rate, thanks again for your help, Lee. I look forward to further collaboration.

Harry

[Infinite]

Hi Lee,

Thank you for your encouragement. As the saying goes, when you're inside your own head, you're behind enemy lines. Discussing issues with folks who share a common interest is a lot more efficient!

Regarding your questions: here's a ebay item (I hope it's still up when you check it out, the bidding ends soon) for an old 4x5 enlarger:

http://cgi.ebay.com/Beseler-4x5-photo-darkroom-enlarger-extras-/380359955745?pt=LH_DefaultDomain_0&hash=item588f385521

Although the analog market for photos has shrunk dramatically, there are still large format enlargers in production. Here's a link to a discussion that lists most of the decent manufacturers and prices:

http://www.largeformatphotography.info/forum/showthread.php?t=3557

Here's a link to a larger LED that I mentioned (100 W vs. 30 W):

http://www.ledssuperbright.com/30w-led-c-18/100w-white-high-power-led-p-226

Regarding your work with projecting IR noise, what about using a high power IR LED in a photo enlarger? I imagine you'd have to change the focus a bit (because the lens wasn't designed for that short a wavelength), but in theory, you could get the sharpness and the contrast you need, as well as enough radiant energy (I think that's what they use to measure "brightness" in IR)

Here's a link to a IR LED manuracturer: http://www.ledtronics.com/Products/ProductsDetails.aspx?WP=1737

On another note, how did you get those 12 shots of your first 360degree model without a lot of movement from the model? Was she on a rotating platform?

At any rate, thanks again for your help, Lee. I look forward to further collaboration.

Harry

Interesting links, thanks for sharing Harry. That LED looks mighty powerful!

Thanks for your tips on IR. It's a pain as I really need high contrast with Pattern Projection, so a Digital Projector is ideal. It's also hard to over come the soft focus I get when using other filters in front of the cameras lens. As the DSLR's are Quartz modified.

That's correct I used a computer controlled platform. It rotates slowly so some movement can creep in. Not ideal but better than installing a 24 or 36 Camera system. You would need multiple PC's dozens of wires, a 36 Camera Breeze license, lighting issues etc.

I'm still finding ways to automate the clean-up process. Non-Rigid Registration is the way to go.

Hope to see some test examples from you soon.

Lee

[hsmith1960]

Hey Lee,

I've tried to come up with some solutions to the problem you mentioned of the lack of non-rigid registration methods.

Have you thought of some of the plug-ins (I'm not sure if that's the correct term) for MATLAB? Here's one that seems to fit the bill, at least in its description:

http://www.mathworks.com/matlabcentral/fileexchange/21451-multimodality-non-rigid-demon-algorithm-image-registration

MATLAB is about $1900, which is outside my range. But I understand that Octave, a freeware clone, is available that will run most MATLAB files:

http://www.gnu.org/software/octave/

I signed up to download the MATLAB algorithm (even though I haven't purchased the main software), but I haven't downloaded the Octave package yet to see if everything works.

Just wanted to keep you updated

Harry

[Infinite]

Hi Harry, thanks for looking further into it. I don't think that is really suited to 3D Meshes.

There are only a few studios in the world currently, who have their own Non-Rigid code. It's very expensive and varies from situation to situation.

The main guy behind allot of the research is called Hao-Li.

There is nothing really publicly available at the moment Hopefully there will be soon.

[Infinite]

A new update on the Scanner-Killer 6 Cam system. 2 tier Capture.

0.2 seconds to capture per slice, 8 slices needed for a 360. About 10-20 seconds a full 360 Capture. 1-2 Million Polygons per Capture Slice using the new Mega Pixel algorithm from xyz-rgb

A whopping 36MP texture resolution in full colour, per Capture slice.

Very little post-processing in ZBrush.

Still more improvements to be made but I think I am finally getting some where.

Megapixel Captures (3 Slices yet to be merged):

http://www.ir-ltd.net/uploads/SK-Megapixel-01a.jpg

36MP Compared to older 10-12MP Capture:


http://www.ir-ltd.net/uploads/SK-MP-Comparison-01a.jpg

Can't show the colour textures yet as this is still under-wraps as I am working with another CGFeedback member on this. Hope to share all this new workflow and how these are captured in the next few weeks.

Setting up and calibrating a 6-8 Camera System is pretty tricky. Allot to handle.

[hsmith1960]

Lee,

Very impressive results! I'm looking forward to more details.

BTW, I'm still working on the software for non-rigid registration. I don't have much experience with Octave (Matlab clone) coding, so it's going slowly.

Harry

[Infinite]

Lee,

Very impressive results! I'm looking forward to more details.

BTW, I'm still working on the software for non-rigid registration. I don't have much experience with Octave (Matlab clone) coding, so it's going slowly.

Harry

Thanks Harry, I'm really interested to see how you get on. I don't think it will be easy to find a solution. Nearly every major Visual Effects studio (who rely on scanning) and every major Scanning and 3D Capture Developer need a Non-Rigid solution. Some have their own, some desperately need it. So it is in high demand.

[james_blonde]

For those worried about the flash being sufficient to 'wash out' the projected noise pattern, here's a solution from a photographer (me):

Use a neutral density filter on the 3rd camera to subdue ambient light, and turn up the flash power to compensate. The more ND you use and the more you turn up flash power, the less projected pattern or other ambient light will be visible.

[Infinite]

For those worried about the flash being sufficient to 'wash out' the projected noise pattern, here's a solution from a photographer (me):

Use a neutral density filter on the 3rd camera to subdue ambient light, and turn up the flash power to compensate. The more ND you use and the more you turn up flash power, the less projected pattern or other ambient light will be visible.

Hi Graham, it sounds like an interesting idea. I would be intrigued to see it in practice as you have to use a very bright projected patten.

The other issue is having to use a 3 camera per pod system, which can be prohibitive and costly. Just using 2 is far easier to manage and scale up to like 8 or 16 cameras.

I will certainly try that technique out though, thanks for sharing.

[Infinite]

For anyone interested in Stereo photogrammetry processing, xyz-rgb have just release their off-line version of Scanner-Killer for $299.99

Which means no more processing on the cloud, you can achieve far superior results to Photofly or Agisoft using this app, as long as you follow the set-up and calibration guidelines well.

More details - http://www.scannerkiller.com/store.html

http://www.scannerkiller.com/pdf/Scanner%20Killer%20V1.05%20Specs.pdf

[mala]

Just got the email from them,great news....but no Mac version!:mad:

[Infinite]

For those worried about the flash being sufficient to 'wash out' the projected noise pattern, here's a solution from a photographer (me):

Use a neutral density filter on the 3rd camera to subdue ambient light, and turn up the flash power to compensate. The more ND you use and the more you turn up flash power, the less projected pattern or other ambient light will be visible.

Hi Graham, I tried the ND filters. No good. They hardly did anything. It's very difficult to get rid of the pattern because it is so bright. I tried various ND filters and Variable versions. No joy :confused: The ideal scenario is using InfraRed projection but there are a host of issues associated with this.

Merry: Perhaps blast them an email with a Max port suggestion? Or even better .....buy a PC clap

[james_blonde]

Hi Lee, the ND filters will work (assuming your flash is strong enough) - it's just a matter of how much you need. Without knowing how bright your projected pattern is, it is hard to say how much you will need, but if you add 8 stops of ND filter and increase the flash intensity by 8 stops, then you will have reduced the projected pattern intensity to 1/256 of its original value. That may or may not be enough.

Another approach would be to try using polarized light. If you projected the pattern through a polarizing filter and then add a separate polarizing filter to the camera, you should be able to rotate one of the filters until the pattern disappears from the camera's viewfinder.

[Infinite]

Hi Lee, the ND filters will work (assuming your flash is strong enough) - it's just a matter of how much you need. Without knowing how bright your projected pattern is, it is hard to say how much you will need, but if you add 8 stops of ND filter and increase the flash intensity by 8 stops, then you will have reduced the projected pattern intensity to 1/256 of its original value. That may or may not be enough.

Another approach would be to try using polarized light. If you projected the pattern through a polarizing filter and then add a separate polarizing filter to the camera, you should be able to rotate one of the filters until the pattern disappears from the camera's viewfinder.

Hi Graham, thanks for the reply. Well I tried a variety of exposure settings, I just couldn't seem to get rid of the pattern. The Projector is an Optoma, 3000 Lumens, 2000-1 Contrast Ratio. I'm also using Venus Interfit 300w flash heads, they don't have a great deal of control on them but they are very bright.

The polarized filter technique sounds good, although I don't quite think it will work. I can certainly give it ago. If it does work, fantastic!

[mala]

Merry: Perhaps blast them an email with a Max port suggestion? Or even better .....buy a PC clap

Guess you mean a Mac port :)...i did mail them but haven't got a reply..
Buy a PC?...Hmmm:rant: I guess if there comes a time when i really need SK,then i'll have to move to the dark side:sw:

[Infinite]

Guess you mean a Mac port :)...i did mail them but haven't got a reply..
Buy a PC?...Hmmm:rant: I guess if there comes a time when i really need SK,then i'll have to move to the dark side:sw:

Sorry yup I meant a Big Mac. Not Max.

I will ask about it, maybe it might be featured in the future.

[Infinite]

Another approach would be to try using polarized light. If you projected the pattern through a polarizing filter and then add a separate polarizing filter to the camera, you should be able to rotate one of the filters until the pattern disappears from the camera's viewfinder.

I tried using Circular Polarizing filters. No joy :( It's really odd, if you hold the Polarisers in your hands and cross them over at the right angle while looking at a Monitor or even in front of the Projected light you can block it out. It goes dark.

Yet when using one in front of the Projector and one in front the Camera, nothing. No matter what angle I set them to all it does is filter out the specularity from the Projection, not the light projection itself. So I think that technique is no good.

[Infinite]

Perhaps Linear Polarizers are the way to go?

[NextDesign]

Hi Harry, thanks for looking further into it. I don't think that is really suited to 3D Meshes.

Actually, you might be able to use it. Think of the image slices that are generated during an MRI. You could possibly use those slices to use 2d-nonlinear registration algorithms, then use some freely available software to reassemble those slices back into full meshes.

Another way you could do it, would be to create some joints along your object, say a head for example, smooth-bind it, and write a simple program that iteratively moves each joint to minimize the distance between the vertex sets of both meshes.

[gjpetch]

No matter what angle I set them to all it does is filter out the specularity from the Projection, not the light projection itself. So I think that technique is no good.
Yeah, the polarization of light is lost and randomized once it's reflected from a diffuse surface, it's only maintained in specular reflection, so it's only going to filter out the specular reflection, regardless of whether it's linear or circular polarization.

[Infinite]

Yeah, the polarization of light is lost and randomized once it's reflected from a diffuse surface, it's only maintained in specular reflection, so it's only going to filter out the specular reflection, regardless of whether it's linear or circular polarization.

Thanks Greg, yeah I found this out the hard way.

My new way around these issues is, I have a company in the US building some LCD Shutters that pick up a signal from a PocketWizard II. This is then able to very quickly block the light from the projector, then reset. They should arrive next week.

[KlaNDeR]

Hi Lee,

Thanks for this fantastic post, your work and effort is unbelievable.

I'm interested in scanning very small objects (no more than 10cm). Do you think that scanner killer can work with two dslr with macro lens and pico projector?

Thanks!

[Infinite]

Hi Lee,

Thanks for this fantastic post, your work and effort is unbelievable.

I'm interested in scanning very small objects (no more than 10cm). Do you think that scanner killer can work with two dslr with macro lens and pico projector?

Thanks!

Thanks KlaNDeR, yes that should be possible just make sure you have an appropriately sized calibration card to calibrate with. I haven't tried this myself though.

[KlaNDeR]

Thanks KlaNDeR, yes that should be possible just make sure you have an appropriately sized calibration card to calibrate with. I haven't tried this myself though.

Thanks Lee! I hope I can try it out soon.

[morbo]

Hey Lee,

any results from those LCD shutters you mentioned?

It sounds like an interesting idea.

[morbo]

Hey Lee,

I was just looking at your free head that you posted at the following link:
http://www.ir-ltd.net/infinite-3d-head-scan-released

Awesome stuff!

and its at a level of quality that I'm aspiring towards, but I'm not sure how to get there with my existing setup.

and am hoping you could help out.



using a 2 camera (canon 60d w/50mm 1.8's and an optoma projector) setup and a projector. Unfortunately, I wasn't able to get anything usable without the projector.

The attached picture is of a portion of a mesh generated by SK core and my setup.

As you can see that its a little "pixelated" for lack of a better word. I know that there are smoothing settings built in, but my concern is the lack of details I'm able to extract out of this mesh without a considerable amount of work. I'm guessing that this is why its worth it to buy the full SK package.

That said, how can I make this into something closer to your model?
You mentioned that you use zbrush to project your scan mesh onto a mesh with better/proper topology?

Can I assume that you do a fair bit of post work on the model to make it usable? Or is it simply, I"m using the wrong version of SK if I want high end meshes?


Thanks Lee, you're a godsend in the world of 3D scanning and stereophotogrammetry!

[Infinite]

Hey Lee,

any results from those LCD shutters you mentioned?

It sounds like an interesting idea.

Yeah thanks the testing went very well, it's a good solution.

Hey Lee,

I was just looking at your free head that you posted at the following link:
http://www.ir-ltd.net/infinite-3d-head-scan-released

Awesome stuff!

and its at a level of quality that I'm aspiring towards, but I'm not sure how to get there with my existing setup.

and am hoping you could help out.



using a 2 camera (canon 60d w/50mm 1.8's and an optoma projector) setup and a projector. Unfortunately, I wasn't able to get anything usable without the projector.

The attached picture is of a portion of a mesh generated by SK core and my setup.

As you can see that its a little "pixelated" for lack of a better word. I know that there are smoothing settings built in, but my concern is the lack of details I'm able to extract out of this mesh without a considerable amount of work. I'm guessing that this is why its worth it to buy the full SK package.

That said, how can I make this into something closer to your model?
You mentioned that you use zbrush to project your scan mesh onto a mesh with better/proper topology?

Can I assume that you do a fair bit of post work on the model to make it usable? Or is it simply, I"m using the wrong version of SK if I want high end meshes?


Thanks Lee, you're a godsend in the world of 3D scanning and stereophotogrammetry!

Yeah you are correct I do a fair amount of post work, especially on the final sculpt. Scanning is only really useful for base reference to work with, steal from. I honestly didn't know the released SK Core version had such a hard polygon limit I think it will make it hard for people to evaluate properly. I would say get used to Core first before thinking about upgrading. I also recommend checking out other applications as well to evaluate.

It is also possible to get OK results without a Projector but sometimes it is pot luck. The calibration, focus and lighting is key to get good extraction.

Unfortunately I'm having to curb the amount of information I share now because other companies in the UK are starting to offer the same kind of work/service that I offer and ideas tend to "leap" from studio to studio if you see what I mean. But I hope the start of this thread might offer some insights that might help your research.

I will say however, this type of capture really is not for the faint-hearted. It is a long (expensive) road to go down to get good results, I have been at this for about 2/3 years now, blood tears and sweat and I am still not quite there yet.

[morbo]

Thanks for all the info Lee,

I'm on the other side of the pond, so no competition, and plus...I only have sk core ;)
================================================== =============
Off the sk forum:

The SK CORE version ($399) does not allow you to up-res the quality of the 3D extraction. It is capped at 130,208 extracted polygons or less.

As a workaround you could calibrate to a smaller window and photograph your object in sections to generate a higher combined resolution.

================================================== ===========

It makes it more difficult to work with the data, but on the other hand....I guess it protects those like yourself who have invested in the full version.

Kinda sucks....but "c'est la vie"

I'm kinda waiting for kinect fusion to become more of a reality, if I'm going to go lowrez, then I might as well do it quickly and cost efficiently.

As for other software, Agisoft? seems like it would be the only affordable solution ....other than a structured light setup..

Though, I really liked the results out of the nextengine3d scanner...just slow in comparison (different beast).

Anyway, thanks for the info....I've been on this for about a year now....its expensive...and time consuming and painful at times...with very little return at the moment.
Peace :)

[Infinite]

I agree with you, Kinect is the one too watch for sure. Also take a good look at the other softwares out there. I wouldn't recommend NextEngine personally, poor customer service I found when dealing with them.

Keep at it, you will get good results.

Lee

[NextDesign]

I wouldn't recommend NextEngine personally, poor customer service I found when dealing with them.
Really? I had no problems with them while I was working at my old job.

The software's a bit buggy, but it can provide some great results.

As for other software, Agisoft? seems like it would be the only affordable solution ....other than a structured light setup..

For active scanning, a friend of mine has created his own 3d scanner, and is looking to start selling it at the beginning of next year: http://www.russian3dscanner.com/

For passive, you could try using Bundler (http://phototour.cs.washington.edu/bundler/) and CMVS (http://grail.cs.washington.edu/software/cmvs/), which are both free, but only give point clouds which would have to be meshed; or Photofly (http://labs.autodesk.com/utilities/photo_scene_editor/), which I've found to be great.

[Infinite]

Really? I had no problems with them while I was working at my old job.

It was related to general business matters and the person I dealt with was quite blunt and rude, which put me off them totally. Never good for business. Also the scanner looks pretty limited

For active scanning, a friend of mine has created his own 3d scanner, and is looking to start selling it at the beginning of next year: http://www.russian3dscanner.com/

Andrew (Neo) has posted on here, his work looks ace. Can't wait to see it released soon.

[morbo]

I agree with you, Kinect is the one too watch for sure. Also take a good look at the other softwares out there. I wouldn't recommend NextEngine personally, poor customer service I found when dealing with them.

Keep at it, you will get good results.

Lee


Thanks Lee!

[Kel Solaar]

Cool video: http://www.youtube.com/watch?feature=player_embedded&v=zHEi55oJJOA :)

[Infinite]

Most excellent, thanks for sharing Kel.

[meshmasters]

Man, that's sweet!

[Kel Solaar]

They should have called it MiniDebevecRig :) Looks like the results are very clean!

KS

[Infinite]

They should have called it MiniDebevecRig :) Looks like the results are very clean!

KS

Or better yet ...MiniUSCICTRig I'm sure more than one pair of hands, minds and $grants$ went into those previous builds !

[tonytrout]

I see that Microsoft is selling its Kinect for Windows stuff now.

http://blogs.msdn.com/b/kinectforwindows/

[oglu]

some interesting surface scans...
http://www.fxguide.com/featured/new-source-for-3d-displacement-textures/

[Kel Solaar]

some interesting surface scans...
http://www.fxguide.com/featured/new-source-for-3d-displacement-textures/

That looks fantastic!

KS

[Infinite]

some interesting surface scans...
http://www.fxguide.com/featured/new-source-for-3d-displacement-textures/

Cool link, some interesting Rig iterations there! thanks for sharing Christoph.

[oglu]

interesting...
http://www.cgarena.com/archives/news/digital-double_journey2.php

[Infinite]

Very interesting, although hard to know how good the capture data is, without seeing any scans? Good to see the Lightstage 6 IS being used, I heard it was defunct and gathering dust...