Support NeoGAF

[gofreak]

ah, now it's revealed! gpgpu for pseye! sure... but could happen, of course. would still be worse than dedicated hardware, though (and worse than fpga, that is).

More resource intensive. But not necessarily 'worse', quality wise, than a hardware implementation.

Stereo reconstructed depth maps are noisier than ToF data, but is it 'good enough' for 'good enough' skeletal tracking?

SoftKinetic's library is based on depth maps, so I presume they are doing the requisite stereo reconstruction on PS4 in a performance envelope that we can surmise is - at least - acceptable for a game like Just Dance 4.

 

[szaromir]

More resource intensive. But not necessarily 'worse', quality wise, than a hardware implementation.

Stereo reconstructed depth maps are noisier than ToF data, but is it 'good enough' for 'good enough' skeletal tracking?

SoftKinetic's library is based on depth maps, so I presume they are doing the requisite stereo reconstruction on PS4 in a performance envelope that we can surmise is - at least - acceptable for a game like Just Dance 4.

Has Sony even shown skeletal tracking?

 

[c0de]

More resource intensive. But not necessarily 'worse', quality wise, than a hardware implementation.

Stereo reconstructed depth maps are noisier than ToF data, but is it 'good enough' for 'good enough' skeletal tracking?

SoftKinetic's library is based on depth maps, so I presume they are doing the requisite stereo reconstruction on PS4 in a performance envelope that we can surmise is - at least - acceptable for a game like Just Dance 4.

Yeah, that's what I meant with "worse". It will be expensive to do so on gpu.
Just read about "Delaunay triangulation" on gpu and it seems it will need quite some resources. And this is with a gpu that is also to be used to render images.
I don't say it won't suffice for games but the tech itself is different to kinect right from the start and differs in building internal images, processing (needs) and outcome.

 

[satam55]

saved your a** in the discussion? not really

Nope! I already had read through this thread before: http://m.neogaf.com/showthread.php?t=452355

So, I already knew GPGPU would be a big part of the PS4 Eye Camera.

 

[Tripolygon]

ah, now it's revealed! gpgpu for pseye! sure... but could happen, of course. would still be worse than dedicated hardware, though (and worse than fpga, that is).

General purpose hardware is not better than dedicated hardware. News at 5.

 

[Futureman]

My roommate got the camera. Played around with Playroom... Other than that, what is the use?

He also got 4 controllers and realized Knack co-op is his only multi play game ha.

 

[Tripolygon]

sure ...like the infinite power of the clouds

Yea there is definitely something wrong with you. I don't argue with crazy people so go away.

 

[gofreak]

Has Sony even shown skeletal tracking?

No. But it seems there is a third party middleware option at this point. I do not know if Sony intends to provide their own library or leave it to middleware.

 

[c0de]

Nope! I already had read through this thread before: http://m.neogaf.com/showthread.php?t=452355

So, I already knew GPGPU would be a big part of the PS4 Eye Camera.

You are basing this on what *could* be and present it now as a fact?

 

[c0de]

General purpose hardware is not better than dedicated hardware. News at 5.

You brought this into this thread, not me. So live with appropriate replies or go away.

 

[Tripolygon]

You are basing this on what *could* be and present it now as a fact?

Fact

1. You don't need full body 3D skeletal tracking to play Just Dance, the software need only recognize your hands and feet. Just like it does on PS4 using software.

2. Stereo camera is better at gesture recognition and finger tracking than ToF cameras

3. ToF camera has higher latency than stereo camera tracking

4. Sony has been at this camera tracking thing and augmented reality for ages

5. Kinect is a more hardware based solution for Z depth recognition while Stero Camera is a software based solution. So we can deduce that Kinect would be better at it than PS4 Eye.

You brought this into this thread, not me. So live with appropriate replies or go away.

Yes i did, go look up what OpenCV and what OpenCL can do. You don't need an FPGA to get a reliable depth map from stereo camera setup. And it is absurd to suggest that.

 

[szaromir]

No. But it seems there is a third party middleware option at this point. I do not know if Sony intends to provide their own library or leave it to middleware.

It's just amusing that there are people arguing how much better PS4 Eye is to Kinect with nothing to show for it and are using some random PC demos as their examples. It's obvious the device is an afterthought for Sony, they definitely shy away from showcasing it.

 

[Alx]

And that is where GPGPU comes in, shocking i know.

GPGPU won't magically fill the holes. If there is no way to differentiate pixels of the same surface, you cannot estimate depth by stereovision alone. You'll need higher level iterations to extrapolate data where it cannot be measured, and that will always be less accurate than a direct measurement.

Fact

1. You don't need full body 3D skeletal tracking to play Just Dance, the software need only recognize your hands and feet. Just like it does on PS4 using software.

Actually for a game like just dance, silhouette extraction is a better approach. It can still be matched to a silhouette template when incomplete, while estimating the exact position of specific points (hands and feet) is more difficult.

2. Stereo camera is better at gesture recognition and finger tracking than ToF cameras

Not true. At best stereovision could more easily offer a faster framerate than ToF, but there's nothing inherent to that approach that makes finger or gesture tracking easier. If anything, finger trackign is harder with stereovision because of the self-occlusion of fingers, which is less of a problem when you have a single point of view.

3. ToF camera has higher latency than stereo camera tracking

Stereo camera tracking adds the depth map reconstruction step, and if the result isn't accurate it can make the tracking more complex. If anything there are chances that latency would be higher.

4. Sony has been at this camera tracking thing and augmented reality for ages

Sony always chose the "easy path" for their computer vision algorithms, hence the Move and his glowing bulb. Their image processing demos were at best "state of the art", never really groundbreaking.

 

[kalmeq]

Ok ps4 camera is not kinect 2 here something from ps4 camera review:

''Lighting conditions appear to play a role in how well it will recognize who’s trying to log in. If I have the overhead lights on, it will log me in almost instantly every time, but if I am using lamps, daylight, or light sources aside from those used to get the facial recognition data, I am hard pressed to be able to successfully log in this way.''

Kinect can see you no matter what is lighting condition not the ps4 camera.

''Impressive features such as eye tracking are being worked on (some in partnership with NASA) and could drastically alter this from being a simple, yet impressive technical demo device to being a real cutting edge tool in technological entertainment advancement.''

no eye tracking which kinect has.

source:http://www.*************************/2013/11/18/playstation-4-camera-review/

 

[c0de]

Yes i did, go look up what OpenCV and what OpenCL can do. You don't need an FPGA to get a reliable depth map from stereo camera setup. And it is absurd to suggest that.

Using FPGAs has different intentions but it's not to get something "reliable" in this context.

 

[Arttemis]

When did people start embracing motion supplemental gestures as being so essential to their experience? Do people really enjoy the Kinect beyond the Just Dance-powered house party?

Also, what motions are people expecting to perform that would be unrecognizable by some easily implementable middleware on the PS4? We know it's capable of tracking hands, feet, and heads, as well as gestures - all at the speed of the Kinect or better, on top of voice commands.

 

[B_Boss]

what trying..... i never see a single video of a ps4 doing body recognition (calculating the distance where it is..im not talking about face recognition)..playroom that is the techdemo of the camera capabilities work with the controller exactly how it would work on the ps3 the move...but now have a pad and is inside the controller....and is clearly more precise...

the triangulation can be done via software and it would take enormous resources or maybe someone is looking in that specs also hardware triangulation?

The Just Dance 2014 commercial no?, unless you're going to preempt its authenticity.

just stop selling the ps4eye for something that isnt...it wont be a kinect ...it wanna be a psmove + (and maybe is better for it)

Who the hell ever tried "selling the ps4eye for something that isnt..." lol? Sony (relatively speaking) hardly ever spoke about their Camera, and it could be that they obviously know it isn't as technically 'advanced' as their competitor's Camera but then again The XB1 is essentially built around the Kinect (at least its my opinion) whereas Sony's approach is that it is strictly a peripheral that is not absolutely essential to the experience anyone could have on their console (again, relative to their competitor).

 

[VanWinkle]

sure ...like the infinite power of the clouds

You are a joke here, and your opinions are both often incorrect and should not be taken seriously.

Anyway: To answer your question or assertion that it doesn't do triangulation.

 

[Alx]

''Impressive features such as eye tracking are being worked on (some in partnership with NASA) and could drastically alter this from being a simple, yet impressive technical demo device to being a real cutting edge tool in technological entertainment advancement.''

no eye tracking which kinect has.

Kinect doesn't have eye tracking either : eye tracking means tracking the fine motions of the pupils, which requires to be very close to the camera (see the demo in the case of Sony, but it wouldn't be different with kinect).
But it could still be done the same way on kinect or on PSEye (actually more easily with kinect since eye tracking is much easier in pure IR images), it's nothing really hard to do when you're close enough. I'm not sure it would be very useful in a living room, though.

 

[Tripolygon]

Using FPGAs has different intentions but it's not to get something "reliable" in this context.

What makes a good tracking game.

1. Head tracking Check

2. Hand tracking Check

3. Face Recognition Check

All these but kinect is better is just stating the fucking obvious.

 

[c0de]

What makes a good tracking game.

1. Head tracking Check

2. Hand tracking Check

3. Face Recognition Check

All these but kinect is better is just stating the fucking obvious.

Yeah, good response to my statement. Mission accomplished.

 

[Skeff]

Anybody pointed out Kinect 2 is 30fps but PS camera operates at between 60 and 240fps.

 

[Alx]

2. Hand tracking Check

That's a very, very basic tracking that wouldn't fly as a real product. It's probably based on the OpenCV sample of histogram backprojection, that will track any element with a given color distribution in the scene (and that has the bad habit of including in the tracked object random elements of the environement). Notice how the background is perfectly white, and it tracks the forehead of the user in the beginning.

 

[B_Boss]

You are a joke here, and your opinions are both often incorrect and should not be taken seriously.

Anyway: To answer your question or assertion that it doesn't do triangulation.

@Kayle: dat assertion

 

[TheKayle]

You are a joke here, and your opinions are both often incorrect and should not be taken seriously.

Anyway: To answer your question or assertion that it doesn't do triangulation.

ppl are throwing everything on the gpgpu capability of the ps4....the infinite power of the gpgpu..lol

u was asking audio? gpgpu... fluid simulations? gpgpu.. body tracking? gpgpu..ai offloads gpgpu.....it relaly sound like the infinite power of the clouds

 

[B_Boss]

ppl are throwing everything on the gpgpu capability of the ps4....the infinite power of the gpgpu..lol

u was asking audio? gpgpu... fluid simulations? gpgpu.. body tracking? gpgpu..ai offloads gpgpu.....it relaly sound like the infinite power of the clouds

By god you sound like misterxmedia buddy (I don't assume you're him but the choice of words and logic are astounding). Most astute folks I know of seem to praise the PS4's architecture in a positive light and rightly so based on observation, evidence and the pedigree of those involved who designed it. Relative to its competitor the GDDR5 option is convincingly a better choice, especially concerning the developers who will use the hardware and software set of said consoles.

If the GPGPU can (hypothetically) actualize the functions you have mentioned above then it can actualize them. I don't see many here exaggerating but stating facts and at the very least, trying to be accurate.

 

[Tripolygon]

That's a very, very basic tracking that wouldn't fly as a real product. It's probably based on the OpenCV sample of histogram backprojection, that will track any element with a given color distribution in the scene (and that has the bad habit of including in the tracked object random elements of the environement). Notice how the background is perfectly white, and it tracks the forehead of the user in the beginning.

I know that is a very basic tracking. I have seen more impressive demonstrations like PlayRoom for example which comes preinstalled on PS4 has a better hand tracking demo in there.

 

[satam55]

By god you sound like misterxmedia buddy (I don't assume you're him but the choice of words and logic are astounding). Most astute folks I know of seem to praise the PS4's architecture in a positive light and rightly so based on observation, evidence and the pedigree of those involved who designed it. Relative to its competitor the GDDR5 option is convincingly a better choice, especially concerning the developers who will use the hardware and software set of said consoles.

If the GPGPU can (hypothetically) actualize the functions you have mentioned above then it can actualize them. I don't see many here exaggerating but stating facts and at the very least, trying to be accurate.

LOL! He does, doesn't he?

 

[Skeff]

ppl are throwing everything on the gpgpu capability of the ps4....the infinite power of the gpgpu..lol

u was asking audio? gpgpu... fluid simulations? gpgpu.. body tracking? gpgpu..ai offloads gpgpu.....it relaly sound like the infinite power of the clouds

not only can it do that, but you complain about this in a camera thread probably without realizing that Kinect is confirmed to be using GPGPU for body tracking....hence the 10% reserve.

 

[Alx]

I know that is a very basic tracking. I have seen more impressive demonstrations like PlayRoom for example which comes preinstalled on PS4 has a better hand tracking demo in there.

Is it the menu navigation part ? I don't think it requires hand tracking either, you can just detect a motion flow in the area above the controller, without knowing if it's a hand or any other object.

 

[Tripolygon]

Anybody pointed out Kinect 2 is 30fps but PS camera operates at between 60 and 240fps.

Yes but it is very reliable and accurate. ToF and Stereo camera's work differently, so for stereo camera to get good tracking you need faster fps to prevent the camera from losing track of the object while Kinect tracks infrared lights. Very basic explanation but it should work.

 

[TheKayle]

not only can it do that, but you complain about this in a camera thread probably without realizing that Kinect is confirmed to be using GPGPU for body tracking....hence the 10% reserve.

im not complain nothing and its clear that gpgpu can help.....im smiling at the fact the tppl seem to throw everything in the gpgpu dream like if it dont have to deal with tradeoff

ppl expecially sony fanboys...are bshitting that 10% reserverd all the day long....and
knowing how triangulation work....and the problem that triangulation with two cameras without infrared have....it would need a lots of reasource for a complete body track...expecially if encounter problem as....more ppl with same dress colors..or background with the same color of the dress..ecc ecc

infrared is fundamental for body tracking..and not just for light problems

 

[ShapeGSX]

2. Stereo camera is better at gesture recognition and finger tracking than ToF cameras

Please tell me how you came to this conclusion. Is it just from that one video of a close up of a single hand on a uniform, grey background?

 

[le.phat]

ppl are throwing everything on the gpgpu capability of the ps4....the infinite power of the gpgpu..lol

u was asking audio? gpgpu... fluid simulations? gpgpu.. body tracking? gpgpu..ai offloads gpgpu.....it relaly sound like the infinite power of the clouds

how you arn't banned is just boggling my mind. You must be joke account of one of the mods. Has to be.

 

[le.phat]

By god you sound like misterxmedia buddy

I've seen him post on there, so it's no suprise.

 

[Dictator93]

ppl are throwing everything on the gpgpu capability of the ps4....the infinite power of the gpgpu..lol

u was asking audio? gpgpu... fluid simulations? gpgpu.. body tracking? gpgpu..ai offloads gpgpu.....it relaly sound like the infinite power of the clouds

You may be one of the most blindly biased posters on GAF. Just throwing that out there.

 

[Sleeplessnights]

Is there any game that uses the controller and camera to move around the split screen?

 

[casmith07]

So....does anyone have the camera?

Thinking of buying it.

 

[Tripolygon]

Please tell me how you came to this conclusion. Is it just from that one video of a close up of a single hand on a uniform, grey background?

I came to this conclusion because kinect as it currently is can only detect a closed and an open hand. I figured they could use the 1080p camera sensor to track the digits better but all they have talked about is Kinect knows the difference between a closed palm and an open palm.

 

[TheKayle]

You may be one of the most blindly biased posters on GAF. Just throwing that out there.

im not payed from ms nor from sony...ps4eye capability (hw and sw) cannot be compared to what kinect is doing......thats it and most ppl...(most sony lovers) find this hard to accept..i understand it

ps4eye is a kinect1+ but i suspect is more a psmove++ ....nothing more than this...

dual camera expecially without ir...in trackin is not good...at all

this mean that the kinect is fantastic? no ..the kinect 1 was horrible....i bought the psmove and not the kinect in fact

 

[szaromir]

When did people start embracing motion supplemental gestures as being so essential to their experience? Do people really enjoy the Kinect beyond the Just Dance-powered house party?

Also, what motions are people expecting to perform that would be unrecognizable by some easily implementable middleware on the PS4? We know it's capable of tracking hands, feet, and heads, as well as gestures - all at the speed of the Kinect or better, on top of voice commands.

Why are you trying to downplay it? From the user (player) perspective motion (gesture or full body) controlled games and UI need to "simply work", when the user has to accommodate when the device is going to work and when not then the experience falls short. MS seems to have addressed many of the original Kinect's limitations that should make using it simply work, whereas Sony has yet to show the proof of that (and it seems that by design it's not going to be reliable in dimly lit rooms with it using RGB sensors).

 

[satam55]

I've seen him post on there, so it's no suprise.

Really??!!!! Do you have a link?

 

[GrizzleBoy]

I came to this conclusion because kinect as it currently is can only detect a closed and an open hand. I figured they could use the 1080p camera sensor to track the digits better but all they have talked about is Kinect knows the difference between a closed palm and an open palm.

What is quite ironic, is that the first video you showed in this thread showcasing the middleware Sony is using was the best case of hand/finger tracking (MUCH better than what you showed in this stereo camera video representing PS4 Eye tech), except that they were using a time of flight camera (Kinect Tech) in your first video.

Barring all this, PS4 Eye is just not equal to Kinect.

Being reliant on light sources for depth accuracy and having zero of the IR blasting tech means that the Eye doesn't have the same abilities both inside and outside of gaming.

 

[jmdajr]

well it's on 60 bucks right?

 

[TheKayle]

how you arn't banned is just boggling my mind. You must be joke account of one of the mods. Has to be.

as i say everytime i meet ppl that find hard to accept other persons opinions....

ignore me ...and pls stop to communicate with me if your msg have to be insults or blatan cry


ps for the misterxmedia story..tell them also when on semiaccurate i told him to stop dreaming and realize that the xb1 was a 1.2 tf console

 

[ShapeGSX]

I came to this conclusion because kinect as it currently is can only detect a closed and an open hand. I figured they could use the 1080p camera sensor to track the digits better but all they have talked about is Kinect knows the difference between a closed palm and an open palm.

That's just because they are tracking people in a wide space like a living room. If they aimed the camera right at your hand, and were just a couple feet away like in the video you linked to, I'm certain that Kinect would do better, and in more varied environments (not a grey background).

You know, something like this:
http://www.youtube.com/watch?v=xqyfWWlAnVI

That's just not what Microsoft was aiming for.

 

[Tripolygon]

That's just because they are tracking people in a wide space like a living room. If they aimed the camera right at your hand, and were just a couple feet away like in the video you linked to, I'm certain that Kinect would do better, and in more varied environments (not a grey background).

You know, something like this:
http://www.youtube.com/watch?v=xqyfWWlAnVI

That's just not what Microsoft was aiming for.

Now show me where I said it couldn't do it.

 

[ShapeGSX]

Now show me where I said it couldn't do it.

I didn't say that you said it couldn't do it.

You said, "Stereo camera is better at gesture recognition and finger tracking than ToF cameras."

That is the assertion I was challenging. I believe that your statement is wrong.

 

[cjtiger300]

Well, you can't blame him for not being a native english speaking person...

Yeah my bad. I shouldn't judge. I would look a fool trying to type in Italian.

 

[Tripolygon]

GPGPU won't magically fill the holes. If there is no way to differentiate pixels of the same surface, you cannot estimate depth by stereovision alone. You'll need higher level iterations to extrapolate data where it cannot be measured, and that will always be less accurate than a direct measurement.

Of course it is less accurate, it is a software solution, that much is pretty obvious.

Actually for a game like just dance, silhouette extraction is a better approach. It can still be matched to a silhouette template when incomplete, while estimating the exact position of specific points (hands and feet) is more difficult.

And that is my point, you do not need full body skeletal tracking or kinect to play just dance.

Not true. At best stereovision could more easily offer a faster framerate than ToF, but there's nothing inherent to that approach that makes finger or gesture tracking easier. If anything, finger trackign is harder with stereovision because of the self-occlusion of fingers, which is less of a problem when you have a single point of view.

Actually it is true, you would have to be right up close to kinect for it to recognize separate digits.

Stereo camera tracking adds the depth map reconstruction step, and if the result isn't accurate it can make the tracking more complex. If anything there are chances that latency would be higher.

It is more complex because it is a purely software based solution. I am not arguing otherwise but it works.

Sony always chose the "easy path" for their computer vision algorithms, hence the Move and his glowing bulb. Their image processing demos were at best "state of the art", never really groundbreaking.

I think you mean cheapest path. ToF was created because it was easier to do than using software with stereo cameras.