Tips for using Skanect with Structure Sensor


#1

Here are some of my ideal settings for getting the best results with Skanect.

Settings
Feedback Quality = Medium
Offline Recording = All Frames
Force QVGA = OFF
Track Loss Detection = ON
Depth = Depth & Color
Uplink Color Gain = Locked on Record

Process
Watertight = Low/Medium
Colorize ->
Resolution = 2/3 mm
Inpaint Colorless = ON
Highest Quality = ON
Face Mode = ON (Person) otherwise OFF

Create a config.ini file with the line (center-y-as-height-percent = .75) and upload it to “Prepare -> New -> Config file”
(This will allow you to start scanning at someone’s face as opposed to their torso 75% height level) Don’t forget to change it if you want to change your default starting point.

The best scans that we have seen use bright but diffused artificial (not sunlight) lighting from multiple sources to illuminate the subject and eliminate shadows. What I have done in the past is attach 1 or 2 of these lights to the back of my iPad with velcro when I am going to be at an offsite location with unknown light sources.

You want your Bounding box to be just a bit bigger than your subject but not overly big or else you will saturate the results. For instance if I wanted to scan a standing person I would typically set the following:
Bounding box = 1.1 x 2.2 x 1.1
Aspect ratio = Height x 2

If the person is taller than 6’ you may need to increase the box size

If I just wanted to get head and shoulders I might use.
Bounding box = 0.6 x 0.6 x 0.6
Aspect ratio = Normal

When you are starting a scan of a person start the capture at the face first take one step to the left to capture the side of the nose then continue to the right and around the subject in a spiral fashion (similar to a tornado). I would not recommend going back to the person’s face again if you can avoid it.

You may need to make a sweep up the back at the end and do a top down view in order to capture the top of the head.

You want to aim to get a continuous 30 FPS in Skanect when scanning. Some great things to try. Ensure Skanect is the only program running. Disable your firewall or antivirus to see if that helps to speed things up.

Finally, if you have the ability to invest in a CUDA 2.0 or higher NVIDIA GPU this will also help to get better results.

Hopefully you will find you will get better results by following the above steps. Once you get into the swing of things you will soon adapt to your own preferences. Here’s a sample file for you to see using Skanect 1.8 + Structure Sensor + iPad Air 2.

Good luck!


Scanning an object by rotating it
Need some advice on proper usage
Scanning app limitations
#2

Hi. Thanks for the reply. We will try the settings you suggested to see if we get any kind of improvement. We are running a Quadro K2200 (CUDA 5.0?) The PC is high end as it spec’d for CAD Design. So I don’t think we have any hardware limitations causing the problems. Just to alleviate equipment concerns here is our setup:
Here is our configuration:

IPAD Air 2 (Model MGKL2LL/A - Version 9.3.2)
Structure Sensor (Firmware 2.0 - SN#36597)
Asus RT-AC3200 Router
PC Running Skanect is hardwired to the network
IPAD Connected via Wifi on 5-2 Channel. Dedicated 5-2 Channel to IPAD Only.
PC Specs running Skanect:
Win 7 64Bit
I7 3.66Ghz
16Gb RAM
Nvidia Quadro K2200 (4GB GDDR5)


#3

I use a very similar process to your yours except that I run color at highest resolution. I’m still not sure what Uplink Color Gain actually does. I also turn off Track Loss Detection. I put my subject on a rotating turntable and that helps me get more consistent results, especially in regards to lighting. I highly recommend the Arqspin tables. I’ve got the 24" table which is only supposed to hold up to 200 lbs, but I regularly scan people heavier than that and it’s lasted for over a year.


#4

@davidb that setup looks pretty solid to me. You should be getting great results with that.


#5

@barkerjeremyscott
Uplink Color Gain
Locked on Record (RECOMMENDED) = Sets the color exposure based on the first frame (I’d use this in almost all cases especially where you have questionable lighting, shadows etc)

Automatic = adjusts exposure throughout the scan based on your current viewpoint (I’d use this if I had great studio quality lighting and if you just want to see if you can get a better resulting scan)

Track Loss Detection
ON (RECOMMENDED) - This should usually be set as you will get help if you lose tracking.

OFF - If you are using a turntable OR if you are a Skanect expert and know how to prevent the loss of tracking.


#6

I think you have left info out regarding the config.ini file. When we setup as you suggested we get a huge bounding box and we are only trying to do the head. So when you say create an .ini file it now has no option for bounding box. Bounding box = Config.ini and there is no way to change.

Using the above setting you provided and doing a head only like in my previous post is probably one of the worst scans we have done to date.

So I guess you can see why there is frustration because it seems like you need to do trial and error with every setting until you get something resembling a decent scan.


#7

We have not gotten 1 good scan since we purchased in September 2015. So we are going on a year now waiting for hi-res scans and hoping that eventually they would fix. We even sent our whole setup (Air 2 + Sense Scanner) and there was no improvement.


#8

The bounding box and config file are two separate things.

re: Config File
Create a file called config.ini and within that file, include, for example, the following line:

      center-y-as-height-percent = .75

Upload the config.ini file via Prepare -> New -> Config file
This allows you to start at 75% up from the bottom as opposed to the normal default of 50%.

re: Bounding Box
I would start with the following and adjust accordingly.

      Bounding box = 0.6 x 0.6 x 0.6
      Aspect ratio = Normal

re: Your Sensor
I remember your sensor and personally testing it. We found no issue with your sensor or the resulting scans.

Skanect is more focused on geometry as opposed to texture. When using Uplink we are limited on the amount of data we can pass through so we must pick and choose. We choose performance and geometry over texture.

See the scan I uploaded above at the start of this thread. Does the resolution meet with your expectation and are you able to create a similar scan? (Please note I removed the auto-lighting from Sketchfab, this can make a big difference in appearance). If the resulting scans as shown do not meet your expectations you may want to look into in a different sensor such as this one.


#9

Bounding Box said Config File. However, did not realize that you could click on it and change to bounding box.

Tried another scan but Structure says Beta Over and cannot open it.

Also what you are saying about Geometry vs. Color you do realize you answered my post in the Beta Forum and moved to the general forum. And you do realize a big new feature of the 1.9 is HI-Res Color Scanning.

So I will continue this over in the beta forum.


#10

All signs point to Apple’s TestFlight default expiration kicking in and disabling the app for current External Testers included on the distro list. Uploading another build for distribution shortly.


#11

Tried to create a config.ini file with just the line you suggested. My goal is to start the scan at 75% height of the box. Didn’t work.
Could you please post your config.ini file so I can try it?


Will Skanect Ever Get an Update? Is this a Dead Product?
#12

@info9 Sure here you go.
config.ini


#13

So what ever happened to the 1.9 update for Skanect? This forum never mentions the version of Skanect being used. The update is supposed to contain the hi res color images that the iPad is
capable of. This thread appears to be dependent upon version 1.9 but no mention of it is
made. Crystal even suggests using a different scanner (twindom) instead of Structure sensor!

The development team’s focus seems to be on the new version of structure and software to support it. Where does that leave those of us who have been using Structure Sensor for 2 years and making our voices clear that we have needed better resolution for textures and waiting for the release of Skanect that can take advantage of the iPad’s camera quality.

When will Skanect version 1.9 be released to see if it fixes davidb’s issues?


#14

@bob2 streaming higher resolution color data over wireless networks isn’t a simple upgrade. We’re working hard to not just make this work, but to make it work at the level of stability that already exists with Skanect 1.8. In the meantime, we’ve launched 1.8.4 to address some immediate upgrades, but our team is still working very hard to launch 1.9.

We appreciate your patience.