Near mode explained

[Joshua Blake]

This post has more details on Kinect for Windows near mode and what they did to make it work better than the Kinect for Xbox sensor.
http://blogs.msdn.com/b/kinectforwindows/archive/2012/01/20/near-mode-what-it-is-and-isn-t.aspx

Josh

[Γιάννης Γράβεζας]

"Additionally, in version 1.0 of the runtime, near mode will have some skeletal support, although not full 20-joint skeletal tracking (ST).  The below table outlines the differences between the default mode and near mode:"

And then continues to say that the only "sceletal" joint that will be tracked will be the center hip joint, aka something akin to the geometric center of the object. So let me rephrase:

"Additionally, in version 1.0 of the runtime, near mode won't have skeletal support, but there's some good ole blob tracking for ya all.  The below table tries to make us seem smart and you dumb. Thanks for all the fish:"

PS Joshua, would it be possible for you to arrange a hangout with the K4W developers?

--
bliss is ignorance

[So Townsend]

Blob detection implies the algorithm.

They probably are using the RTF with less labels which allows the trees to be processed by less intensive methods.  

Without GPU assistance, it is very hard to create many reliable joint proposals in real time applications.  It is much easier to do a few.  Specifically, the chest, head, and shoulders are easily processed because of their distinct shapes.

I assume they don't want to deploy a product only capable of being run on newer computers with real shaders.  Or risk the problems with taxing a good CPU with joint tracking.

Cheers

-So Townsend

Sent from my iPhone

[Kyle McDonald]

> As some have observed, the Kinect for Xbox 360 sensor was already technically capable of seeing down 50 centimeters - but with the caveat "as long as the light is right".

looks like they're reading this mailing list ;)

so it sounds like near mode is something genuinely different in the
kinect firmware, not just a software level change. what would be
really great to see is a comparison of depth images, from the
different cameras, showing the spotty data for XBK then the good data
from K4W

kyle

2012/1/21 Γιάννης Γράβεζας <wiz...@gmail.com>:

[Γιάννης Γράβεζας]

2012/1/21 So Townsend <so.to...@gmail.com>

Blob detection implies the algorithm. 

They probably are using the RTF with less labels which allows the trees to be processed by less intensive methods.  

Without GPU assistance, it is very hard to create many reliable joint proposals in real time applications.  It is much easier to do a few.  Specifically, the chest, head, and shoulders are easily processed because of their distinct shapes.

I assume they don't want to deploy a product only capable of being run on newer computers with real shaders.  Or risk the problems with taxing a good CPU with joint tracking.

Cheers

-So Townsend

Sent from my iPhone

Still, one point per object != skeletal tracking. That part was plain old indecent marketing speak. Did you thought the column labelled "Other 19 joints" was amusing or/and insulting to your intelligence? 

 

--
bliss is ignorance

[Joshua Blake]

2012/1/21 Γιάννης Γράβεζας <wiz...@gmail.com>

Still, one point per object != skeletal tracking. That part was plain old indecent marketing speak. Did you thought the column labelled "Other 19 joints" was amusing or/and insulting to your intelligence? 

 

 

Just because you don't like it doesn't mean it's marketing speak or insulting your intelligence. Not everything is a conspiracy trying to trick you!

 

The player indexes are not just blob tracking. The player indexes are an artifact of the skeleton tracking algorithm and only track human forms, not any random blob that moves.

 

My understanding is that in near mode, the accuracy of the skeleton tracking with all 20 joints is not yet at the same level as in default mode. In order to ship by Feb 1st, they decided to only report the player indexes and hip center joint, which is a partial result of full skeleton tracking. That's a better solution than reporting bad data.

 

I and others have requested K4W add skeleton tracking in near mode as well as partial skeleton tracking (upper body, etc.) and I believe they are looking at that for a future release.

 

PS Joshua, would it be possible for you to arrange a hangout with the K4W developers?

 

What do you mean for hangout? Are you in the Redmond, WA area? I could ask if they'll do an online Q/A session with the OpenKinect community via voice or chat, and I'm sure they would like to do that.

 

Josh

[Γιάννης Γράβεζας]

2012/1/21 Joshua Blake <josh...@gmail.com>

2012/1/21 Γιάννης Γράβεζας <wiz...@gmail.com>

Still, one point per object != skeletal tracking. That part was plain old indecent marketing speak. Did you thought the column labelled "Other 19 joints" was amusing or/and insulting to your intelligence? 

 

 

Just because you don't like it doesn't mean it's marketing speak or insulting your intelligence. Not everything is a conspiracy trying to trick you!

Look, Microsoft doesn't have a good reputation, they've repeatedly used unethical practices against the open community. Every move they make triggers the "conspiracy" sensors on lots of people, not just me. That's not my fault, karma is a bitch, and it will take more than sweet talking to reverse that situation. That doesn't have to do with the particular post though. It's the way that they presented a deficiency that was just silly. Why would they need to use a table to tell you about the lack of the "Other 19 joints", couldn't they just say "near mode will only support 1 joint for now". That was pure marketing spin 

 

 

The player indexes are not just blob tracking. The player indexes are an artifact of the skeleton tracking algorithm and only track human forms, not any random blob that moves.

 

My understanding is that in near mode, the accuracy of the skeleton tracking with all 20 joints is not yet at the same level as in default mode. In order to ship by Feb 1st, they decided to only report the player indexes and hip center joint, which is a partial result of full skeleton tracking. That's a better solution than reporting bad data.

 

I and others have requested K4W add skeleton tracking in near mode as well as partial skeleton tracking (upper body, etc.) and I believe they are looking at that for a future release.

 

It's their component labelling algorithm that actually indexes the blobs. They have to isolate distinct features from the frame to perform skeleton tracking on each one of them 

 

PS Joshua, would it be possible for you to arrange a hangout with the K4W developers?

 

What do you mean for hangout? Are you in the Redmond, WA area? I could ask if they'll do an online Q/A session with the OpenKinect community via voice or chat, and I'm sure they would like to do that.

 

A voice chat would be optimal. I live in athens greece though(EET timezone) so please take that in consideration if/when you manage to arrange that. Thanks 

 

Josh

--
bliss is ignorance

[Jonathan Gennick]

2012/1/22 Γιάννης Γράβεζας <wiz...@gmail.com>

Why would they need to use a table to tell you about the lack of the "Other 19 joints", couldn't they just say "near mode will only support 1 joint for now". That was pure marketing spin 

 

I don't see spin in that table. The author is emphasizing what he sees as key information. If I just eyeball the article without reading the body text, I can still come away knowing that near mode gives me depth mask and hip tracking. The figure and the table seem to do a pretty good job of summarizing what's really important to know. 

I don't mind the lumping together of the other 19 joints. Keeping the table to three columns means that it fits nicely on the page. 

Jonathan Gennick
http://gennick.com/bio.html
Brighten the corner--where you are!

2012/1/22 Γιάννης Γράβεζας <wiz...@gmail.com>

[Γιάννης Γράβεζας]

2012/1/23 Jonathan Gennick <j...@gennick.com>

2012/1/22 Γιάννης Γράβεζας <wiz...@gmail.com>

Why would they need to use a table to tell you about the lack of the "Other 19 joints", couldn't they just say "near mode will only support 1 joint for now". That was pure marketing spin 

 

I don't see spin in that table. The author is emphasizing what he sees as key information. If I just eyeball the article without reading the body text, I can still come away knowing that near mode gives me depth mask and hip tracking. The figure and the table seem to do a pretty good job of summarizing what's really important to know. 

I don't mind the lumping together of the other 19 joints. Keeping the table to three columns means that it fits nicely on the page. 

Jonathan Gennick
http://gennick.com/bio.html
Brighten the corner--where you are!

Yeah but all that information could be neatly provided in a single line of text, bolded so the eye catches it. Of course by going that route, they'd have to admit that they won't provide any actual/usable skeleton tracking for near mode. That would be the most sincere approach considering that this post was intended for developers who'll have to deal with a half baked product on launch. If a little spin makes it easier for them to swallow, I guess it's none of my business to even mention it, especially since I don't intend on using the SDK. Sorry about that and I really didn't meant to discredit the K4W development team. I know it's hard in there 

Yannis

--
bliss is ignorance

[Joshua Blake]

Your complaint is about the formatting of a blog post. Are you done now? You've added nothing to this conversation.

---

From: Γιάννης Γράβεζας
Sent: 1/23/2012 9:31 AM
To: openk...@googlegroups.com
Subject: Re: [OpenKinect] Near mode explained

[So Townsend]

Why are you people still talking about this...

"they'd have to admit that they won't provide any actual/usable skeleton tracking for near mode."

Again, the system should technically be capable of supporting skeleton tracking.   Because of hardware restrictions, Microsoft has not rolled the system with full skeleton tracking.  They most likely will as soon as APUs become more common and the pathetic excuse that intel calls a graphics card is no longer in the majority of computers.

Also, remember that Microsoft is making k4w for the atypical c# developer.  No disrespect meant to that community ofcourse, many have families to take care of and can't spend months hacking with a complicated API.  

Sent from my iPhone

[Γιάννης Γράβεζας]

2012/1/23 So Townsend <so.to...@gmail.com>

Why are you people still talking about this...

"they'd have to admit that they won't provide any actual/usable skeleton tracking for near mode."

Again, the system should technically be capable of supporting skeleton tracking.   Because of hardware restrictions, Microsoft has not rolled the system with full skeleton tracking.  They most likely will as soon as APUs become more common and the pathetic excuse that intel calls a graphics card is no longer in the majority of computers.

But it does provide skeleton tracking for the default mode, doesn't it? The problem is that they where in a rash to provide a great new feature so they could have an excuse for the inflated pricetag. They came up with their "near mode" which they had to admit was already there. So, basically it's the same old kinect with some minor fixes that won't end up on the xbox kinect of course. And they want to sell those minor fixes at double the price of the same product which is already in the market. Kinect for windows is a marketing rip off. It hurts, but it's the truth and it must be said

 

Yannis

--
bliss is ignorance

[Jonathan Gennick]

2012/1/23 Γιάννης Γράβεζας <wiz...@gmail.com>

The problem is that they where in a rash to provide a great new feature so they could have an excuse for the inflated pricetag.

I admit, the pricing rubbed me wrong at first. Then I thought more about it and realized that Microsoft was between a rock and hard place. They were caught out and unprepared with a business plan for the non-xBox market. 

I like to think that it if were me, that I would have kept the price the same and just accepted the fact that not every Kinect would map to a game-buying xBox owner. Some "leakage" in the business plan is acceptable, right? But that is too easy to say when one's job is not on the line over making his or her numbers.

I do believe that in the long run, that Microsoft would do well to differentiate the products. Make  a lesser version at a lower price aimed at the xBox. Make a higher version that will appeal strongly to PC users. Remove any artificial restrictions and rely upon the feature differences to drive most buyers in the desired direction. Some small number of people will buy the cheaper device and use it with a PC, but let that go by and don't worry about it. That is the approach I would aim for in the long term, but who knows what Microsoft will do. 

[Γιάννης Γράβεζας]

2012/1/24 Jonathan Gennick <j...@gennick.com>

2012/1/23 Γιάννης Γράβεζας <wiz...@gmail.com>

The problem is that they where in a rash to provide a great new feature so they could have an excuse for the inflated pricetag.

I admit, the pricing rubbed me wrong at first. Then I thought more about it and realized that Microsoft was between a rock and hard place. They were caught out and unprepared with a business plan for the non-xBox market. 

I like to think that it if were me, that I would have kept the price the same and just accepted the fact that not every Kinect would map to a game-buying xBox owner. Some "leakage" in the business plan is acceptable, right? But that is too easy to say when one's job is not on the line over making his or her numbers.

 

It is the accumulation of bad decisions by folk driven from the need to "make their numbers" that has brought our civilization to the brink of collapse. Numbers are tools and meant to be used by humans, not the other way around

 

I do believe that in the long run, that Microsoft would do well to differentiate the products. Make  a lesser version at a lower price aimed at the xBox. Make a higher version that will appeal strongly to PC users. Remove any artificial restrictions and rely upon the feature differences to drive most buyers in the desired direction. Some small number of people will buy the cheaper device and use it with a PC, but let that go by and don't worry about it. That is the approach I would aim for in the long term, but who knows what Microsoft will do. 

No one can know for sure, but my bet is that they'll do the blunder stunt next. Smells like fresh vista coming out of the oven

 

Jonathan Gennick
http://gennick.com/bio.html
Brighten the corner--where you are!

--
bliss is ignorance

[Γιάννης Γράβεζας]

This is an excerpt from a recent thread in the K4W forum (http://social.msdn.microsoft.com/Forums/en-US/kinectsdk/thread/1fb914e9-e294-4f58-92d9-ae95583d96c1):

Whoops, sorry I should have formulated more clear. I am only referring to development for windows, either with the old "Xbox" kinect and the new "Microsoft" kinect.
 
OK, so if the answer is yes, then I still don't get the difference from the two devices.
 
@Todd Ferkingstad: from this http://blogs.msdn.com/b/kinectforwindows/archive/2012/01/20/near-mode-what-it-is-and-isn-t.aspx blog post:
 
"This resulted not only in the need for new firmware, but changes to the way the devices are tested on the manufacturing line. "
 
Can't you just update the firmware on the old kinect then?(This guy's asking wrong questions, notify the goons) 
 
"The lenses on the Kinect for Windows sensor are the same as the Kinect for Xbox 360 sensor, so near mode does not change the field of view as some people have been speculating."
 
So if there's no increase in range but just better ability to focus, what's different (hardware-wise)?(Well there's... and maybe... bah, send the goons already)
 
Please be patient as I truly dont understand. (That's cause you're dumb and we're smart)

I guess the K4W team is about to evolve from a development department to damage control and reputation management. Fresh vista anyone?

PS That guy wasn't me, the grumpy know-it-all. It was an atypical c# developer who has a family to take care of and can't afford to waste money on marketing frauds. Cheers