Resolution by Skanect

[cs hui]

Hi, all well experienced Skanect members:

 

I finally setup a new laptop to do 3D scanning using Kinect and Skanect.

My Laptop configuration is :

OS :  Window 10 64-bit

CPU :  Qaud core 2.7 - 3.5 GHz

Memory: 8GB

Graphic:  Nvidia GeForce 940M with 2GB memory

Hard drive :  1TB @7400rpm

When I started Skanect, I set the Feedback Fidelity to High,  the frame rate is only 6 - 8 fps.  When I set the Feedback Fidelity to Medium, the frame rate is about 27 - 30fps, but in the log file, there are a lot of error messages saying  "number of images not written to disk because of overload....". 

I used the MSI performance screen to track the performance of the machine:

CPU utilization 20-30%, GPU utilization 30 - 60%

Memory utilization 20 - 30%

Disk utilization 10 - 20%

It seems to me my laptop can handle the job easily.  Now the situation is that no matter how I set the Feedback Fidelity, either I got low frame rate or high frame rates but with a lot of frames not recorded.  I suppose this will affect greatly on the resolution of the final 3D object produced by Skanect.

Any suggestion from anyone please!

C.S.

[Tobby Ryan]

I have a quadcore i7 laptop with 8gb ram as well.My video card is an intel HD3000, therefore unsupported by skanect. I can however capture in medium res to get 12fps.  So i capture in medium, and do an offline reconstruction in high.  i cannot even use GPU construction.

The main difference is, whilst capturing and displaying in realtime high isnt possible on my laptop due to the 5-8fps in high, i have to not do a realtime scan. i capture the data and show the realtime in a lower resolution.this keeps me from dropping as many frames. i then go back and do the offline reconstruction (as it captured more frames) to see the resolution at high. it will reconstruct back to high resolution and things look great again.  I even had these issues with the Geforce 610 video card unitl i upgraded to a Geforce GTX970 video card in my desktop.  I was upgrading my desktop and did the video card last.  GTX cards have much more ump than non.


I am not sure what kind of answer the Skanect team will give you. Its a trick that i used on my laptop until i started scanning with my desktop strapped to a luggage cart to move around from place to place easily :)

another trick is when you reconstruct, shrink the bounding box to where you don't have extra as much. this will give you a better model as well.

[cs hui]

Hi Tobby,

I did the same as you are doing.  During image recording, I set the Feedback rate to medium and I got nearly up to 30fps.  However, if you study the log, you will find that not many frames are captured because it seems that Skanect has dropped a lot of frames due to "high workload" on disk.  As a result, you cannot got as many frames as you expect to do Reconstruct, no matter what your setting is in Reconstruct.  During Reconstruct, you can see the circle of cameras but at the same time, the circle shows large gaps between cameras.  it is quite noticeable showing that frames have been dropped.  Until this problem is solved, you cannot expect the resolution can improve.  I am still waiting from Skanect's reply.  Please stay tuned.

C.S.

[Vitaly Vyazovsky]

It won't affect resolution much - if at all. Low fps is bad for tracking between the frames, but usually 12-15 is good. 
If you need more resolution, you need to increase the amount of voxels per scanning volume, and right now in skanect it's limited to maximum 512x512x512 - and cannot be altered.
See my post here:
https://groups.google.com/forum/#!topic/skanect/xf9Ug7yRiuQ

суббота, 30 января 2016 г., 1:09:48 UTC+4 пользователь cs hui написал:

[cs hui]

Hi all,

Finally after talking to the Skanect Support many times, the setting for scanning should be as follows:-

Before starting Skanect, turn off any anti-virus program.  This will avoid many frames being dropped due to disk activity too busy.

In body scan, the settings for recording should be

1.  Feedback Quality = Medium (to reduce scanning time and the chance of getting double images on the face)

2.  Force QVGA = Off (It should be the default value unless your scanner cannot support it)  The maximum  resolution is 640x480

In still object scan, the settings for recording should be

1.  Feedback Quality = High (still object would not move and no need to worry about double images)

2.  Force QVGA = Off (It should be the default value unless your scanner cannot support it)  The maximum  resolution is 640x480

In Reconstruct -> Fusion

Fidelity = High

Use the buttons on the right hand side to reduce the box as small as possible.

In Process -> Fill holes

Smoothness = Low