Log in or register to post comments

Why iPad 2 outperforms newer devices

October 13, 2015 - 4:34pm #1

Hi there.

My project is an app that requires recording physical distances from the camera.  In doing so I've discovered large errors in the calibration of newer (retina?) devices (in another thread).  It also means I've been able to compare quite a few devices head to head - and to my surprise the iPad 2 was the clear winner in terms of max tracking distance, physical accuracy, and stability.

Aside from physical distance issues in new devices, I would like to know why newer devices are not killing the iPad 2.  The iPad Air 2 had slightly poorer performance than the iPad 2 which indicates you are not using even a fraction of the full power of these devices.  I totally get that you downsample the texture from camera's frame buffer but it really looks like you are stymying the newer devices.  I would have expected a huge maximum distance from the iPad Air 2 camera....no?  

The following results were using a 15cm image target under ideal indoor conditions.  My definition of stable measuring is that the AR measure had a variance of less than 1-2cm, which I tended to cap at 3m (which was sufficient for my needs).  Max distance was simply measuring the maximum tracking distance regardless of stability.  

Device

Stable Measurement Distance (m)

Max Distance (m)

Camera Resolution

(Megapixels)

IPhone 4

2

3.5

5

iPhone 6

3

~5

8

iPad 2

3

6.8

0.92

iPad Mini

2.5

4.8

5

iPad 4

3

~6

5

iPad Air 2

3

>6

8

 

 

MODE : MAX_QUALITY

Vuforia 5.05

Unit 5.2

OS: various iOS 7.1 - 9.02

 

So my question is, am I better moving to something like ARToolkit where I have fine grain control over the AR pipeline?

Thanks

Scott

 

Why iPad 2 outperforms newer devices

October 30, 2015 - 3:44pm #5

Thanks Mike.  

I'll check check again when I get the chance.

In regards the relative performance of the devices I'd just ike to point out that regardless of calibration (which does play its part) you have cameras running with MUCH higher camera resolutions (than the native 0.92 megapixels of the iPad 2) and able to resolve detail at greater distances.  I"m well aware that on higher res cameras that the video texture is downsampled before it goes into the engine proper.  This is for bottleneck reasons of course, but it means that we are losing out on so much potential fidelity and performance .  Using something like ARToolkit I have control over the size of the texture going into the AR "engine".  The answer (in my opinion) is to allow custom camera profiles so a user could override the default one, and say define the (re)size of the video texture.  BUT.....of course you fear opening a can of worms....perhaps make it an unsupported "feature" :)

Cheers

Scott

 

Why iPad 2 outperforms newer devices

October 29, 2015 - 9:51am #4

Hi Scott

The newer iOS devices should have been calibrated recently and uploaded to the servers.  To make sure your devices have downloaded the new calibrations, launch your app with a network connection, allow the app to run for a bit (about a minunte), and then force quit/relaunch the app.  In the second run, the updated calibrations should be loaded.

Thanks

//Mike

Why iPad 2 outperforms newer devices

October 29, 2015 - 9:01am #3

This is probably due to camera calibration.  The iPad2 may just have the most accurate calibation of the devices you've tested.  I'll forward this topic to the folks in charge of calibration and see if there are any updates for these devices scheduled.  Unforatunatly, with the sheer number of devices on the market, it's not possible for us to hand calibrate every single device.

Why iPad 2 outperforms newer devices

October 13, 2015 - 4:36pm #2

Just thinking out aloud here but perhaps downsampling is introducing artefacts where the iPad 2 texture you use native?  Just guessing?

Log in or register to post comments