PS4 SDK 2.0 brings big improvements to Camera functions & more.

GPGPU depth processing, 1.5 ms to compute depth at 160x100 resolution. Yikes! I also hope the images show what I've said before about stereo cameras being lousy at depth extraction and Kinect being far better. XB1's Kinect really is amazing(ly under appreciated) tech
 
Shifty Geezer said:
GPGPU depth processing, 1.5 ms to compute depth at 160x100 resolution. Yikes! I also hope the images show what I've said before about stereo cameras being lousy at depth extraction and Kinect being far better. XB1's Kinect really is amazing(ly under appreciated) tech
Click to expand...

You do realize that's up to like 600 frames per second right?
 
It's 1.5 ms of GPU time. Yes, if all you want to do in your game is get depth info and display it, the lib runs at 600 fps at 160x100 resolution. If you want higher resolution, there'll be in increase in processing time. If that is linear, than 320x200 will take 6ms, one third of your frame time. and linear is likely a best case scenario, hence the 160x100 example (wanting to give devs a low impact option and not scare them off using motion detection with significant GPU resource hogging).

For accurate motion tracking, Move is going to be essential, but Sony have basically ditched Move for PS4. AFAIK there isn't one announced title in development for it. 160x100 res input might be okay for UI interaction, although that'll require a lot of smoothing and added latency, I expect.
 
Nearly 10% of GPU time per frame at 60 fps. At 160 x 100. And the results are not good compared to Kinect.

And now I'm getting sad feels about Kinect not getting used again ... :(
 
Shifty Geezer said:
It's 1.5 ms of GPU time. Yes, if all you want to do in your game is get depth info and display it, the lib runs at 600 fps at 160x100 resolution. If you want higher resolution, there'll be in increase in processing time. If that is linear, than 320x200 will take 6ms, one third of your frame time. and linear is likely a best case scenario, hence the 160x100 example (wanting to give devs a low impact option and not scare them off using motion detection with significant GPU resource hogging).

For accurate motion tracking, Move is going to be essential, but Sony have basically ditched Move for PS4. AFAIK there isn't one announced title in development for it. 160x100 res input might be okay for UI interaction, although that'll require a lot of smoothing and added latency, I expect.
Click to expand...

function said:
Nearly 10% of GPU time per frame at 60 fps. At 160 x 100. And the results are not good compared to Kinect.

And now I'm getting sad feels about Kinect not getting used again ... :(
Click to expand...

The thing is Kinect 2.0 is hardware limited at 30FPS 512×424 pixels on a system that has 500 GFLOPs less GPU power than the PS4 so using 10% of the GPGPU for 160x100 60FPS depth sensing can still result in a better end product.
 
Last edited:
onQ said:
The thing is Kinect 2.0 is hardware limited at 512×424 pixels on a system that has 500 GFLOPs less GPU power than the PS4 so using 10% of the GPGPU for 160x100 60FPS depth sensing can still result in a better end product.
Click to expand...

What?

At equivalent resolution with best case linear scaling. That would be 13.586 ms out of 16.667 ms for 60 fps which would be basically unuseable. Even for 33.333 ms for 30 fps, it would be unuseable at ~40.7% of your available frame time.

Basically you have to use a lower resolution to get get something even remotely as responsive as Kinect v2. I have no idea how you interpret that as a better end product for depth sensing. It will certainly allow you to do something, but nothing as good as Kinect v2. You can perhaps go ever so slightly faster with massively lower resolution. But you'll never be able to match resolution and speed simultaneously or even come close.

Regards,
SB
 
Shifty Geezer said:
GPGPU depth processing, 1.5 ms to compute depth at 160x100 resolution. Yikes! I also hope the images show what I've said before about stereo cameras being lousy at depth extraction and Kinect being far better. XB1's Kinect really is amazing(ly under appreciated) tech
Click to expand...

Not underappreciated, just underused?

I did notice improvements in recognition under lower light in the PS4 menus after the 2.0 update, just like the audio-recognition has improved too. Ubisoft's dancing game also performs much better this year (at least I only tried this years version and it seems to do quite well - you can play with Move, Phone or Camera here, and the Camera mode currently is my favorite - where people basically said last years version was just plain broken).

But yeah, so far it's clear that Kinect 2 would be performing better. It also costs 3x as much right now though, which doesn't help with the being underused part. Also, games like Ubisoft's dancing game don't exactly use the GPU more than 10%, if at all, for graphics ... .

Actually, Sony providing this in the SDK would in fact help Kinect, because there's a (slightly) bigger market that way for games using any form of body tracking.
 
Last edited:
It's better because he's blinded by fanboy vision.
 
Silent_Buddha said:
What?

At equivalent resolution with best case linear scaling. That would be 13.586 ms out of 16.667 ms for 60 fps which would be basically unuseable. Even for 33.333 ms for 30 fps, it would be unuseable at ~40.7% of your available frame time.

Basically you have to use a lower resolution to get get something even remotely as responsive as Kinect v2. I have no idea how you interpret that as a better end product for depth sensing. It will certainly allow you to do something, but nothing as good as Kinect v2. You can perhaps go ever so slightly faster with massively lower resolution. But you'll never be able to match resolution and speed simultaneously or even come close.

Regards,
SB
Click to expand...

Who said that it would have to be at an equivalent resolution to work? it just need to be good enough to create tracking points. some games might only need to keep track of your hand so being able to track your hand at 60FPS could be better than tracking it at 30FPS.
 
onQ said:
Yes I'm a fanboy of tracking at above 30fps for smoother controls.
Click to expand...

Except you wont get nearly the precision that's required for it to actually be smoother. You're ignoring the real data. That's what puts you into the blind faith camp.
 
BRiT said:
Except you wont get nearly the precision that's required for it to actually be smoother. You're ignoring the real data. That's what puts you into the blind faith camp.
Click to expand...


The type of games that will be using motion controls are more than likely going to be games that are not taxing the PS4 GPU so if it came down to it you could have a game that's using 900GFLOPs for rendering & upto 900GFLOPs for tracking if that much GPU time is needed.
 
onQ, you're ridiculous! If the game needs 900 GFLOPS of compute for tracking that is the very definition of motion contol based games taxing the GPU. You're failing to acknowledge the strengths that Kinect has over PS4 cameeras as well. The very fact that you even suggest a gamemay need up to 900 FLOPS for motion control is ridiculous. There are previous threads comparing PS4 cameras and Kinect and I believe you will discover that Kinect is superior in its specific design to capture motion data.

I would like to thank you for this thread though. It's nice to see Sony is working on making its cameras better for depth processing with the use of GPU compute. If it helps more motion control type games flourish then I'm all for it.
 
Sonic said:
onQ, you're ridiculous! If the game needs 900 GFLOPS of compute for tracking that is the very definition of motion contol based games taxing the GPU. You're failing to acknowledge the strengths that Kinect has over PS4 cameeras as well. The very fact that you even suggest a gamemay need up to 900 FLOPS for motion control is ridiculous. There are previous threads comparing PS4 cameras and Kinect and I believe you will discover that Kinect is superior in its specific design to capture motion data.

I would like to thank you for this thread though. It's nice to see Sony is working on making its cameras better for depth processing with the use of GPU compute. If it helps more motion control type games flourish then I'm all for it.
Click to expand...

You really need to go back and read that.



Edit: using 18% of the GPU (about 332 GFLOPs) can process 160 x 100 dots of depth at 120FPS. How am I blind to believe that there will be cases when this will result in a better solution than 512 x 424 at 30FPS?

Even if 320 x 200 dots is needed for tracking they could get 60FPS for around 660 GFLOPs.

Can you really say that 512 x 424 at 30FPS will always be better than 320 x 200 at 60FPS for depth sensing?
 
Last edited:
onQ said:
You really need to go back and read that.



Edit: using 18% of the GPU (about 332 GFLOPs) can process 160 x 100 dots of depth at 120FPS. How am I blind to believe that there will be cases when this will result in a better solution than 512 x 424 at 30FPS?

Even if 320 x 200 dots is needed for tracking they could get 60FPS for around 660 GFLOPs.

Can you really say that 512 x 424 at 30FPS will always be better than 320 x 200 at 60FPS for depth sensing?
Click to expand...

Why you keep saying that Kinect 2 isn't capable to track at 60fps, when we know it can?

http://i.imgur.com/ZDQLITS.jpg

And how can you compare the quality of detailed 512x424p depth image of Kinect 2 which is a direct result of using a top notch hardware solution, with 320x200p depth image that produced by triangulation?

And why Sony used a (or two?) 640x400p image(s) as input to make a 180x100p depth image as output? Is it due to a restrictions on their calculation method or they choose it to reduce the cost of the calculations? Why do you think that they can go for higher quality depth image using more Gflops? What about latency?
 
tuna said:
Doesn't kinect also need gpu power?
Click to expand...

AFAIK, yes it did (or does, depends on how you interpreted MS' decision to ditch Kinnect in the X1 retail box)

Reading the slides, makes you wonder why they didn't just stick a dedicated SoC in the PS4 camera to do some kind of pre-processing to save
on some of the GPU load.

I'm not that happy with the PS4 camera btw. Since a few updates ago, I get the feeling the sensitivty of the microphone has dropped. From
my couch I'm not able to control the PS4 anymore. I have to stand approx 2 meters in front of the mic/TV/PS4.

I can't imagine that there's so much ambient noise in my room. One difference is that the microphone used to be on a shelf sitting 1.5 meters above
the floor. Now it's under the TV at 50 cm height. AFAIK, the mic is at the bottom of the PS4 camera.
 
You must log in or register to reply here.
Back
Top