go back to loat pose, not enough geometry, slow frame rate

[CP HO]

Dear all
 
Sorry for the last partial post.

I wish to start my Skanect project to do a whole body scan

My computer is an i5 4460 with a GT 750 graphic card and 12G RAM.  I bought a turntable from Taobao and the table can make one revolution in one minute.

I asked my model to stand on the turn table, start the scanning and then start the turn table.

There were error messages 'go back to the last pose', 'not enough geometry' and the frame rate was only some 11 per minute.

Its it because of my hardware problem  or was the turntable too fast/

how to increase the frame rate ?  I read that the optimal rate would be some 30 /minute.  Would increase the RAM to 16 G helps or do I need to change to a high end card?

thanks a lot

cpho

[Nicolas Burrus]

Hi,

I assume you meant 11 frames per seconds (the number displayed by
Skanect). This is not very fast, but if your turntable is quite slow,
it should still track ok. Could you post a short video of Skanect
during the process? It would help us diagnose what is going on. Most
likely reasons are:

- The object is too small for Skanect to track its position

- The bounding box includes fixed parts (e.g the support of the table,
which does not rotate with the object)

- The object geometry is not informative enough. For example, Skanect
cannot track a purely planar object. In this case, you can add extra
geometry to the table (e.g small wood blocks or clothes), and remove
them in the postprocessing.

Best,
Nicolas

> --
> You received this message because you are subscribed to the Google Groups
> "Skanect" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to skanect+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[CP HO]

Dear Nocolas

thanks for your email.  I am sorry that I have to reply direct because I cannot post to the forum directly.  My post did not appear in the group. (see attached)

thanks for your suggestions and i shall do a video to you. 

My other question is 

1, How can I increase the frame rate ? I think it shoud be 30 fps

Would it work by changing to a more powerful graphic card?  Connecting to a USB2 port or upgrading the software of the graphic card?  frankly, I am ready to do all of those if needed

thanks

cpho

--

Dr C P Ho

Sola Nobilitas Virtus
(Virtue Alone Ennobles)

[Tobby Ryan]

USB 2.0 port or 3.0 port should be fine for bandwidth concerns. The faster/better the gpu, the more frames per second. I went from an nvidia GT610 to a GTX970, the GTX card gives me 25-30frames per second depending on the subject. I was getting about 12fps in high with the GT card.   Another tactic to get more fps out of a card, if you aren't relying on the live output is to:

Change to medium quality, scan
go to Reconstruct and under fusion, select High.
this will do an offline high reconstruction from the captured data and render a better version than the Medium quality gave you for the live scan.
This is what i have to do with my laptop that doesn't support GPU (older i7 laptop with intel graphics on board)
On my laptop i get 5fps in high, 12 in medium. It was my workaround before i got my desktop upgraded.

[CP HO]

Dear Nicolas , Tobby and all

thanks for the input and advice.

The system did not work for the last few days, had to re-install the windows, the motherboard driver, the graphic card driver, .net 4 framework and the skanect. The system worked finally.

the first experiment went well, the software accepted the frame rate of 12 fps. Can see the model in the kinect sensor windows but the Skanect cannot see the model.

I have taken a video and posted in Youtube.  Would appreciate your advice.  I think it is just the setting problem. Should be close to home after grobbing around for the past 8 months

https://youtu.be/N2lO5KuNJyg

thanks a lot

cpho

[CP HO]

Dear all

I repeated the experiment today.  the skanect can now sense the model - it was green during capture. 

https://youtu.be/dbEJOGw_euw

My problem: 1. very low resolution  2. do not know how to remove the background subject  3. do not know how to set the bounce box.

any advice would be much appreciated

cpho

[Tobby Ryan]

On Wednesday, August 5, 2015 at 9:22:14 AM UTC-4, CP HO wrote:

Dear all

I repeated the experiment today.  the skanect can now sense the model - it was green during capture. 

https://youtu.be/dbEJOGw_euw

My problem: 1. very low resolution  2. do not know how to remove the background subject  3. do not know how to set the bounce box.

any advice would be much appreciated

To change the bounding box, it will be on the first screen under prepare. I use the Height x 2. this allows you to set it with less side space to scan a taller object such as a person. If the table and wall are still in the scan as green, then you might try scanning where objects are not in the way. I tend to keep a good foot to 2 feet on each side of my turntable for clearance so they are not accidentally captured in the live data as part of the model(aka keep bounding box to turntable/model). Static objects that get in scan range with a turntable can be a bad thing.These can either add as garbage to the scan, or play havoc with tracking. Your video shows that you are trying to scan the model but the tracking is set upon the wall and counter as they are big objects as well.

The table and wall are both in the bounding box. The best thing is, do not have anything in Green that you do not wish to capture.
Green = capture   Red = non-capture, such as shiney black vinyl, or too close to the kinect.  Blue = too far away Gray = non-captured.

I also noticed when you are doing a fusion reconstruction, you are not changing from medium to high in the fidelity. The High Fidelity in the fusion section will make the model look better. High Fidelity will take longer but the output will be much better. Also you can shrink the bounding box in the fusion section to be closer to the model to also get more quality.

I have attached a sample of the prepare settings i use.

I hope i have not stepped on any of the Skanect team's toes for responding.
-Tobby

[Nicolas Burrus]

Tobby - this is the community forum, you'll never step on anyone toes, and I'll take this comment as an opportunity to thank you and the other power-users of Skanect for the time you spend helping others!

[CP HO]

Dear Tobby, Nicolas and power users

thanks for the advice,  I shall use the height x2 option in my future experiment.

I repeated the experiment today, frustrated because the sensor can sense the table and door (green) but cannot scan the model in the middle. (did not turn green)

Obiviously the object is not small, I tried to add a book in the turntable but to no avail.

any solution?

[Roel Veldhuyzen]

Your issue here is not the size of the object/person. The issue is that your turntable is too far away. You can see that the part of the wall that registers ends in front of the turntable, it doesn't register behind it.

So you need to move the sensor towards the turntable, or the turntable towards the sensor.

You should also make the bounding box smaller, the walls can not be registered while scanning! if the sensor sees the stationary walls, while it also sees a turning person on the turntable, you will get an unusable result. (Although you could adjust the size of the scanned area later on, you won't get a decent idea of what you actually scanned right away, in this setup :)

So move the turntable, and make the bounding box smaller.

[CP HO]

Dear Roel

thanks very much

Its very frustrating to find the model not scanned while everything SEEMS to be correct,  Now I have the answer

shall do another experiment next week, hope things will turn out fine

cpho

[CP HO]

Dear All

thanks for your help, I am now able to scan and get a good mesh.

the problem is I wish to have a full length model instead of half length.  My question is : can I move the kinect sensor up and down during the process?  (ie horizontal scannign).   the model is put on a rotating table

or instead of moving the sensor up and down, can I TILT the sensor up and down (ie scanning at an angle)

best regards

cpho

[rgblowpr...@gmail.com]

You can move the scanner around ( up, down and tilt up and down ) once the scan starts.

You need to start from half way up the scanned subject ( unless you have changed the staring point in the config file ) and most importantly the scanner point horizontally to the subject. Any angle Introduced at the start will effectively tilt the scan box / area.

The best way to figures these sorts of things out is to have a go and to learn from your mistakes as it won't cost you anything

$0.02

[CP HO]

thank you very much indeed

I shall follow your instruction.  Shall try up and down horizontally and then tilt up and down

there is another method I would try and would be even better (I think).  Take one round with the scanner in a higher position and then another round in a lower position. This is because I have a turn-table which turn automatically.  In other words, I just need to do two rounds instead of one round.

Moving the scanner up and down has one penalty:  the frame rate was around 10 fps, if I move it up and down with the object rotating, it might be too fast for the system.

Anyway will try all the 3 options.  As you said, it cost nothing to try

cp

--
You received this message because you are subscribed to a topic in the Google Groups "Skanect" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/skanect/yvO_X36xoq0/unsubscribe.
To unsubscribe from this group and all its topics, send an email to skanect+u...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[CP HO]

Dear Tobby

How do you select LOW quality scan?

cpho

[Tobby Ryan]

that will be done at the first, under prepare settings, feedback quality

[CP HO]

Dear Tobby

thanks

shall report what I do

cp

[CP HO]

Dear Richard and all

I tried to move the kinect sensor after starting.  However, it seems to me that no matter how I moved, the sensor cannot see the lower part of the body.  ie, upper part green but the lower part black.

the image acquire was of quite good quality, but only half height

Any idea how to get around it?

cpho



​

--
You received this message because you are subscribed to a topic in the Google Groups "Skanect" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/skanect/yvO_X36xoq0/unsubscribe.
To unsubscribe from this group and all its topics, send an email to skanect+u...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Richard Grenfell]

It looks like your bounding box is too small and needs to be double the height.

Follow

Prepare - New - Aspect Ratio - Height x 2

and you should be able to fi