Why is timestamp difference between 30fps and 60fps is almost the same?

[Mona_J_]

So I just figured that the timestamp difference I am reporting into a csv file is the same for both 30 and 60fps. Any clue if there's something wrong or better to say how should I reflect the difference of 30fps and 60fps in my timestamp? also I don't get why sometimes nothing is captured and it outputs only zeros for roll, pitch, yaw and expression_eyes_up score intensity like below. You know how to fix it?

lines of the 30fps:

Timestamp,roll,pitch, yaw, eyes_up
07-06-2015-04:14:35:092,-17.1718101501465,-6.6672625541687,-0.959841251373291,0
07-06-2015-04:14:35:303,0,0,0,0
07-06-2015-04:14:36:284,0,0,0,0
07-06-2015-04:14:36:416,0,0,0,0
07-06-2015-04:14:36:543,0,0,0,0
07-06-2015-04:14:36:676,-10.7574281692505,-6.31971120834351,-6.32161045074463,0
07-06-2015-04:14:36:844,-8.47167873382568,-10.1662149429321,-3.92463374137878,0
07-06-2015-04:14:37:016,-7.62745237350464,-7.35333919525146,-3.70955848693848,0
07-06-2015-04:14:37:184,-6.52208566665649,-9.197340965271,-5.02745676040649,0
07-06-2015-04:14:37:356,-7.83088636398315,-7.54775953292847,-6.28792524337769,0
07-06-2015-04:14:37:488,-9.97107219696045,-4.90489435195923,-1.51094746589661,0

lines of the 60fps:

Timestamp,roll,pitch, yaw, eyes_up
07-06-2015-04:17:06:657,3.85619783401489,18.880163192749,7.99100589752197,0
07-06-2015-04:17:06:788,4.7002592086792,15.5283308029175,8.34747123718262,100
07-06-2015-04:17:06:960,4.59582996368408,15.8914594650269,7.96146821975708,100
07-06-2015-04:17:07:128,4.52651929855347,16.5068550109863,7.92416715621948,0
07-06-2015-04:17:07:299,4.61186552047729,16.4482173919678,7.8523964881897,0
07-06-2015-04:17:07:431,4.73835611343384,16.6897659301758,7.81285047531128,0
07-06-2015-04:17:07:560,4.78607368469238,16.8063697814941,7.7626781463623,0
07-06-2015-04:17:07:691,4.91337871551514,16.5551357269287,7.67645931243896,0
07-06-2015-04:17:07:864,4.89635705947876,16.5885715484619,7.60606956481934,0
07-06-2015-04:17:08:032,4.89934968948364,15.5425539016724,8.15667247772217,0
07-06-2015-04:17:08:164,4.78904438018799,15.9280967712402,8.0695219039917,0
07-06-2015-04:17:08:292,4.58381366729736,16.2226047515869,7.75950288772583,0

[Bradley_F_]

I started looking at this and can say pretty authoritatively, I'd need to see your execution loop code to even begin to hypothesize in a reasonable manner.

Without that information, I can just make general statements like:

-timer-chip timestamps often are not accurate past 100ms.  It's possible there isn't a high resolution timer involved in the generation of that timestamp. (Which is probably not the problem at issue here I'd think.  It would be rather silly for Intel to not know this about their own timer chip.  I assume you are getting those timestamps from the Intel SDK rathe than from a system library)

-The interface doesn't seem to make any guarantees as to detecting faces at a steady frame-rate.  You may be limited by the processing power of your machine.  Perhaps, regardless of how many sensor frames are streaming in to the processing engine a second, your machine is only capable of processing about 5 face recognitions a second from that stream, the way you have it configured at the moment.

But really, without seeing how you are interacting with the API, I can't do anything but throw out unfounded guesses based on my prior experience.

-brad

[daflippers]

-timer-chip timestamps often are not accurate past 100ms.

The system timer will log with 1ms resolution although there may be a quantisation larger than 1ms.    

Given both 30 and 60 fps are logging data at a rate of 6 per second this suggests that either the processing speed or the logging is limited to 6 outputs per second.

David

[Mona_J_]

Here's my code, I am not allowed to upload my files online here so I am sharing the github repo. 

https://github.com/lamiastella/FaceTracking

Also the changes are in this file:

https://github.com/lamiastella/FaceTracking/blob/master/FaceID/FaceID/MainWindow.xaml.cs

Please let me know if you know how to have a more precise timestamp which actually can capture the fps timing.

Thanks,

Mona Jalal.