It is currently Mon Dec 16, 2019 8:02 am



Reply to topic  [ 32 posts ] 
 Sleeping Dogs 
Author Message
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Ever since Grand Theft Auto, there has been a gaming fascination with underworld crime in an open urban environment.  What happens when you take Grand Theft Auto and move it to Hong Kong, China?  You get Sleeping Dogs, a wild game featuring fast cars, Kung Fu smack downs, drug busts, and a regular hankering for Chinese Food (seriously!).


Today, we get to test and review Sleeping Dogs in stereoscopic 3D on AMD's HD3D native format, Nvidia 3D Vision, and DDD's TriDef Ignition drivers.

NOTE: Since AMD's native HD3D does not yet have a means to record stereoscopic 3D screenshots, we were limited to using Nvidia's 3D Vision output to show what the game is capable of (with the exception of DDD TriDef).  The results are nearly identical except for some points outlined in the review below.
Read full article...


Wed Oct 03, 2012 2:09 pm
Profile WWW
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
I made a minor edit to the article and the scores by adding a GG3D bug type. The big problem with this game is that even though there was separation between the left and right view, the actual camera separation (the image source) wasn't far enough. So, if you take the left and right views and overlap them, there are only hairline differences between the two. This isn't the sames as a doubled left image, but it's pretty close.

Regards,
Neil


Wed Oct 03, 2012 4:33 pm
Profile WWW
Cross Eyed!

Joined: Sat Apr 14, 2007 4:34 pm
Posts: 108
Reply with quote
I just downloaded the demo from Steam and did a stereo check. First I thought it was only fake 3D (like Crysis 2) due to the very low separation and resulting minimal depth.
After some research though I found that I was wrong and that the game indeed has perfect dual rendered stereo, at least what I can say based on the levels in the demo.

The trick is to change the games separation/convergence settings in the configuration file above the limits set by the game GUI. Fancy enough after the change the GUI shows the new values and as long as you don't touch the 2 sliders, all other options can be changed without reseting these values within their GUI limits (1...10).

The game settings are located in the XML file

...\Steam\steamapps\common\SleepingDogsDemo\Data\DisplaySettings.xml

<?xml version="1.0"?>

<DisplaySettings>
<Version>1</Version>
<AdapterUID>-1945077602</AdapterUID>
<MonitorUID>71405901</MonitorUID>
<ResolutionWidth>1920</ResolutionWidth>
<ResolutionHeight>1080</ResolutionHeight>
<RefreshRateNumerator>60000</RefreshRateNumerator>
<RefreshRateDenominator>1000</RefreshRateDenominator>
<ScalingMode>default</ScalingMode>
<ScanlineMode>progressive</ScanlineMode>
<Fullscreen>1</Fullscreen>
<Stereo3D>1</Stereo3D>
<StereoEyeDistance>0.029999999</StereoEyeDistance>
<VSync>0</VSync>
<AAQuality>1</AAQuality>
<HighResTextures>1</HighResTextures>
<ShadowRes>2</ShadowRes>
<ShadowFilter>0</ShadowFilter>
<MotionBlur>1</MotionBlur>
<SSAO>0</SSAO>
<FPSLimiter>0</FPSLimiter>
<Option3DDepth>10</Option3DDepth>
<Option3DConvergence>25</Option3DConvergence>

<WorldDensity>2</WorldDensity>
<GlobalSetting>0</GlobalSetting>
</DisplaySettings>

The 2 bold values can be set to get the optimal 3D effect based on the display size and viewing distance. The above values 10/25 are my values for a 50" 3D-HDTV at 1m viewing distance using nVidia 3D vision
15/30 are the values for my sons 27" 3D monitor at 80 cm viewing distance, also using nVidia 3D vision.
The only problem seen so far is the gun crosshair in 2D, but the nVidia laser sight can be used as mentioned in the review.
Also due to the high contrast in the scenes, the game is affected from ghosting especially on my plasma HDTV. Really can't wait for OLEDs to take off...


Thu Oct 04, 2012 3:03 pm
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Can you share some screenshots to demonstrate your results?

Just go here:

http://www.mtbs3d.com/gallery/thumbnails.php?album=40

And click the upload button at the top (make sure you are logged in first!). Once your images are online, copy the stimg BBcode from the gallery to your forum post. Then we will be able to see your 3D images like this:


I would be interested in seeing (based on the same setting) the visual results for on the street/driving, and a cinematic. If your recommendations make a difference, I would be more than happy to revisit this review.

Regards,
Neil


Thu Oct 04, 2012 4:12 pm
Profile WWW
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Does no one have good screenshots to show? I was hoping my review was incorrect and I wasn't setting the game up properly.

Regards,
Neil


Fri Oct 05, 2012 10:04 am
Profile WWW
Cross Eyed!

Joined: Sat Apr 14, 2007 4:34 pm
Posts: 108
Reply with quote
Based on my positive experience with the demo I just bought the game and will post 3D screenshot with my optimized settings soon...
(Currently loading the HD texture pack from Steam)


Fri Oct 05, 2012 11:58 am
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Excellent!

Regards,
Neil


Fri Oct 05, 2012 12:22 pm
Profile WWW
Cross Eyed!

Joined: Sat Apr 14, 2007 4:34 pm
Posts: 108
Reply with quote
So here is the first batch of 3D screenshots from ingame and cutscenes with my optimized settings for a 50" display at 1 meter viewing distance.
All effects/lighting/shadows/rain seen so far are rendered perfectly in 3D space and it really all looks awesome.
My main problem is that this game shows me how bad my plasma HDTV is ghosting wise. But of course I can't blame the game for this.





















Sat Oct 06, 2012 4:47 am
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
I don't think you are far enough in the game yet. Get some driving pictures up too.

Regards,
Neil


Sat Oct 06, 2012 10:55 am
Profile WWW
3D Angel Eyes (Moderator)
User avatar

Joined: Sat Apr 12, 2008 8:18 pm
Posts: 11394
Reply with quote
Wow! That looks really good. May have to check that out.

_________________
check my blog - cybereality.com


Sat Oct 06, 2012 11:31 am
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
So I ran these images through, and they are visually correct. When you get further into the game, there are sequences where you drive a car on the street - that will tell use for sure (the 3D impact was different according to the type of scene it was).

Are you able to have a consistent 3D setting right through? Does the dynamic change between a cinematic and a street walking scene?

Regards,
Neil


Sat Oct 06, 2012 8:12 pm
Profile WWW
One Eyed Hopeful

Joined: Tue Apr 06, 2010 2:20 am
Posts: 49
Location: Norwich, UK
Reply with quote
I have this game and have nearly finished it.

nVidia 3D is fine. GG3D scored it Platinum for me but I gave it 4/5. This was because of the 2D unit target icon and crosshair, though crosshair use is so infrequent it hardly matters.

Cutscenes use the same convergence but the camera often zooms in too close forcing you to either disable the 3D during cutscenes or lowering the convergence (depth in SD). Driving scenes are fine too.

The game also allows toyification, if you don't mind controlling the camera to avoid the unnatural popout.

One of the best looking games in 3D IMO. And very flexible but you need to edit the XML file and be aware the devs got convergence and depth mix up - the big sillies.


Mon Oct 08, 2012 1:23 pm
Profile
One Eyed Hopeful

Joined: Tue Apr 06, 2010 2:20 am
Posts: 49
Location: Norwich, UK
Reply with quote
Sorry forgot to add some images. I couldn't see Nobsi's so soz if they show similar scenes.

I chose a bike and boat.





Mon Oct 08, 2012 1:42 pm
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Hi Andysonofbob,

I'm afraid your images underscore my result with Sleeping Dogs. Take a viewer like sView or stereophotomaker. Now...keep in mind that the separation levels you have here is supposed to push the cameras very, very far apart. If this is working properly, each view should be extremely different from the other.

Take either of your images, and overlap the motorcycle or the boat. What you will see is that while there are minor differences (the HUD separation mainly), the left and right views are largely the same. Yes, they are different camera views (not like the Deus-Ex fiasco), but it's as though no matter what you do with with the separation settings, the actual cameras aren't moving apart further than some fixed setting we have no control over.

If you do the same with Nobsi's images, the differences between the left and right view will be far more pronounced, and there was no requirement to split the scenes as far as what you have here. Something is definitely broken with Sleeping Dogs. The error happens mainly in driving scenes and street scenes - it's not across the board.

I would add that your separation levels are so far part (ridiculously far apart!), I'm wondering if this is a symptom of Sleeping Dog's underlying problem where you might be trying to compensate for something that the game isn't delivering. When looking at Nobsi's images, do you still see a requirement to increase the separation, or is the 3D readily apparent?

Regards,
Neil


Tue Oct 09, 2012 10:25 am
Profile WWW
One Eyed Hopeful

Joined: Tue Apr 06, 2010 2:20 am
Posts: 49
Location: Norwich, UK
Reply with quote
Neil wrote:
Hi Andysonofbob,

Take either of your images, and overlap the motorcycle or the boat. What you will see is that while there are minor differences (the HUD separation mainly)

Regards,
Neil


Sorry dude, I just don't see it. Show me though. I have chosen a view with street lines. Will you line them up and post the image for me please? From what I can see, I see the linear line of the road spreading out evenly in my shots - as you would expect. I had my camera elevated so the convergence point might not be as apparent as it could be. The depth might be a little too high though.

The HUD is 2D. Though I think the minimap is where it should be - I hate HUD elements at the bottom of the screen in depth as it clips the floor! The only issue really is the target icon which is 2D. OK the crosshair is also 2D but my Sleeping Dogs playstyle is hand to hand! :D I didn't even think to use nVidia's laser sights so thanks for the tip!

BTW, I would be curious to see how you have your 3D settings set up in the xml.

Neil, it would be helpful if you would post an example of what you would like to see, with regard to depth perception of a game. If you have in mind an ideal scene from a game you have played, which you felt could make an exemplar of depth perception, take a screenshot and post it please. That would be awesome for people like me.

Hope that's OK. :)


Tue Oct 09, 2012 12:33 pm
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
No problem. I'm just finishing something up, and I'll get it done later today.

Regards,
Neil


Tue Oct 09, 2012 1:17 pm
Profile WWW
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Hi Andysonyofbob,

Ok, this should exemplify the problem and the differences. Please view these images in anaglyph so you can see the results with your glasses off.


The above is one of Nobsi's image samples.


This is the same image realigned so that the character's left/right image are overlapped. Ignoring the HUD elements, it's very clear that there is a distinct left and right view with big differences between them. Remember, it's the differences of the left/right view that makes it look interesting in stereoscopic 3D form. Taking into consideration the amount of separation before the left and right views were overlapped like this, it's clear that the cameras are indeed far enough apart and the images are proportionately different from each other.


This is Andysonyofbob's picture. The separation is even higher than what Nobsi had. In theory, the left and right views should look similarly different from each other.


When we overlap the motorcycle, it's clear that there isn't that much difference between the left and right views. We only start to see differences very far into the scene where it doesn't really matter, and it's not proportionately different when we consider the amount of separation that Andy is showing here. If I'm correct, either the cameras aren't actually separating despite what settings are used in the config (offsetting and separation are two different things), or the convergence results are too radically different from one scene to the next, which is undermining the visual results.

Regards,
Neil


Tue Oct 09, 2012 2:37 pm
Profile WWW
One Eyed Hopeful

Joined: Wed Oct 10, 2012 8:21 am
Posts: 1
Reply with quote
Well it seems that Tridef released a Beta yesterday fixing some of the issues with Sleeping Dogs. I haven't tested them yet, but maybe something to consider if/when updating the review.


Wed Oct 10, 2012 8:26 am
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Thanks to a new beta release of the DDD drivers with related profile fixes, I can now demonstrate my point further.

Here are a series of images before and after alignment. Even though the separation isn't astronomical compared to the earlier images, it should be very clear that they are proportionately different left/right views in all these scenes, and the differences don't just start to happen in far, far into the scene.







Remember that the amount of separation is very low compared to what I showed before, so don't expect HUGE gaps between the left/right views. This is just enough to show that the left/right views are indeed different from each other right through the scene. The last image is a poor example because the two versions aren't that different from each other, but you get the idea I think.

Regards,
Neil


Wed Oct 10, 2012 11:47 am
Profile WWW
One Eyed Hopeful

Joined: Tue Apr 06, 2010 2:20 am
Posts: 49
Location: Norwich, UK
Reply with quote
Neil wrote:
I can now demonstrate my point further.
Regards,
Neil


Dude

I completely take it back. I had to see for myself as I frankly didn't believe it.

I used Fireworks to overlay the two views and Neil is right, there is very little difference in separation. My apologies!

I am disappointed in myself for not noticing. I cannot believe how I was so easily fooled. If I haven't uploaded the image right (likely tbh) and you can't see the superimposed jpg, I could have been playing the game in 2D, with depth; that's how little the separation is.

It did feel 'uncomfortable' and that would be why...

Image


Wed Oct 10, 2012 2:42 pm
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Don't feel embarrassed. I submitted a GG3D profile before the review, and I think I scored it as high as 85%. I was suspicious, but I couldn't test it because the AMD HD3D version didn't have a screenshot ability to test with. Ever since Sniper Elite V2 (which was REALLY embarrassing because it was a 2D+Depth game that I scored higher than I should have before the DDD update), I've been more cautious in how I go about reviewing games. This was easy to miss because at least half the game is rendered properly.

I edited your post to have a thumbnail link rather the the STIMG tag. STIMG is for side by side pictures (cross-eyed is best). The "Thumbnail" option listed in the gallery BBCodes list is for linking to 2D pictures, or for linking to 2D/3D pictures from other forums (STIMG will only work on MTBS because of an HTML5 limitation - not our doing!).

Regards,
Neil


Wed Oct 10, 2012 3:00 pm
Profile WWW
One Eyed Hopeful

Joined: Tue Feb 01, 2011 4:45 am
Posts: 4
Reply with quote
I refuse to play this game without 3d enabled. I just wish the crosshair issue will be fixed.


Mon Oct 15, 2012 12:32 am
Profile
Two Eyed Hopeful

Joined: Thu Dec 27, 2012 4:22 am
Posts: 98
Reply with quote
I've been trying to completely understand the premise of the overlapping images test.
I'm not yet convinced this is a valid test.

Please see this NVidia document:
http://www.nvidia.com/content/GTC-2010/pdfs/2010_GTC2010.pdf

And look specifically at the Parallax Budget section.

As noted, I don't have a complete understanding, but it seems to me that what you are seeing with the boat and motorcycle scenarios is the parallax budget kicking in, because of the infinity in the scene.

If I'm reading this correctly, you cannot have much divergence, if you have pixels at infinity. The best you can expect would be 10%, which to my eye is quite possibly what you are seeing with the overlap tests.


This is also complicated by the need to respect the:
Eye Separation = Interocular / Real Screen Width

Maximum Eye Separation is different per person, but NVidia reports that if it's a close screen an average user cannot go above 50% separation. Hence, the Parallax Budget.


Mon Jan 21, 2013 1:57 am
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Hi Bo3bber,

Thanks for your message. I had a similar debate with John Carmack over Doom 3 BFG because what I found was there was disparity for objects up close, but in the distance, it almost looked as though the objects were just offset from each other. He won that debate because indeed the disparity is only supposed to really happen for up-close objects compared to the rest of the scene.

In the case of Sleeping Dogs, if you compare scenes, there is no reason for the car and the motorcycle to have near-zero disparity. They aren't in the far distance or at infinity. If you compare screenshots between the DDD and Nvidia drivers, it's clear that the disparity is handled very differently (and more effectively) by the DDD drivers in this case.

We should open this up for discussion amongst the members, though. You raise a valid concern in our testing methodology, and this isn't the first time it has come up. Others are welcome to pipe in here.

Regards,
Neil


Tue Jan 22, 2013 4:07 pm
Profile WWW
Two Eyed Hopeful

Joined: Thu Dec 27, 2012 4:22 am
Posts: 98
Reply with quote
Hi Neil,

3D tends to be quite complicated and much more subjective than I'd expect. In thinking about this more, I'm nearly sure that the overlap test is not valid. And, that Sleeping Dogs is doing exactly the right thing here and should not be penalized.

The subjective part is that both you and Andy thought the images looked great when first viewed, but after doing the overlap test you thought it was wrong. To my eye, those images are nearly perfect.

If this isn't the right venue for this discussion, please point me to the right spot. I'm going to try to explain why I think the overlap test is wrong for the Sleeping Dogs case.

Apologies in advance for being NVidia centric, I only have their setup.

--------
First off, when you do the overlap test, you are ignoring the difference in the HUD. What you are essentially doing when overlapping the images is moving the convergence. You can do this easily in the NVidia PhotoViewer, by pressing D and F to move in or out. This is easier for me than editing the files.

It also makes it much more clear that we are simply moving the convergence, not anything more dramatic.

I do not think we can ignore the HUD here, because as we crank the convergence, the HUD pops further and further out of the screen. Even though those far distance elements are more closely aligned, we are dramatically impacting the near elements. The only near element in this specific image is the HUD, which is why it is tempting to ignore it.

You can see this effect, even on the forum page itself. Scroll so that both the motorcycle picture and the aligned motorcycle picture are in view. Switch to the Mono vision mode so they are side by side. Do cross-eyed viewing of both images. You can glance between both. You can directly compare the HUD popping out of screen in the aligned image, and being at screen depth in the first case. These images are small enough that extreme popout is not too taxing.

If you then scroll to the TriDef images, I think you see the same thing- a difference in convergence, which is more satisfying to your eye.

--------
The reason I think the game is doing S3D correctly is based on my reading of that NVidia paper I referenced. That is from a developer conference, where the session was to help developers do S3D correctly. There are several notable things in that PDF.

One is that NVidia specifically recommends that developers NOT allow end users to adjust convergence. This explains some of the confusing locked-convergence games. NVidia recommends that developers handle convergence to meet stylistic or game demands. Just like in this motorcycle case, where they have moved convergence deeper into the screen. On purpose, I would suggest.

The text from that slide:

Quote:
Adjust Convergence
 Convergence must be controlled by the application
 Camera parameter driven by the look of the frame
 Artistic / Gameplay decision
 Should adjust for each camera shot / mode
 Make sure the scene elements are in the range [ Convergence / 2, 100 * Convergence ]
 Adjust it to use the Parallax Budget properly
 Cf Bob Whitehill Talk (Pixar Stereographer) at Siggraph 2010
 Dynamic Convergence is a bad idea
 Except for specific transition cases
 Analyze frame depth through an Histogram and focus points ?
 Ongoing projects at NV



That's a good start for what I'm thinking about when we are talking about the overlap image test. I need to create some images of the slides to talk more clearly about the Parallax Budget.

Slide 41 of that talk is where they talk about the Parallax Budget, and why I think it's a fundamental limitation of S3D that you can have either popout or full depth, but not both simultaneously.


Tue Jan 22, 2013 9:09 pm
Profile
Two Eyed Hopeful

Joined: Thu Dec 27, 2012 4:22 am
Posts: 98
Reply with quote
Ah, interesting. In a very similar talk one year later at Siggraph, 2011, Gateau removes that reference to convergence being controlled by the developer.

http://developer.download.nvidia.com/as ... o_Z-SG.pdf


Tue Jan 22, 2013 9:36 pm
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Hi Bo3bber,

Don't worry about where to and where not to talk about stuff in the MTBS forums. We aren't big on formalities here, and this is as good a place as any.

While we respect the opinions and ideals of DDD, Nvidia, AMD, Sony, and more, MTBS has its own tests and classifications for 3D quality control. So even though these documents are framed as 3D manuals, they are just opinions being stated as fact to help the developers along, and try to establish their own corporate positioning in the market. Even with their best of intentions, they aren't completely factual documents and there is plenty of room for 3D artistry and choice.

Now GameGrade3D isn't about artistic choice. We just want to know what settings need to be turned down to make a game 3D compatible, what visual flexibility there is, and what anomalies and problems remain afterwards. Based on these criteria, we finalize a score. The convergence option isn't treated as a penalty, but there is an indicator as to whether or not the flexibility is there for gamers that want it.

There were some remarks in your message that I need to clarify. I never said I thought that Sleeping Dogs was great in 3D. In the review, I said that Sleeping Dogs' 3D implementation was uneven:

Quote:
This is where things get...unfortunate...for both the Nvidia and HD3D outcomes. There is an inconsistency in the effectiveness of the 3D. As the left and right images are separated, there should be increasingly pronounced differences in how the objects look because both the camera perspectives are changing. For example, even though there is plenty of separation in the above picture between the left and right views, the imagery isn't different enough from each other, and it's these differences that make 3D look interesting!


The remark about Andy's enjoyment of the game is only half true. Yes, Andy enjoyed the 3D experience at first, but then he realized what he was missing! There is a difference there. I enjoyed my black and white TV until I got my color TV and knew what I was missing. ;=)

Ok, now the rest...

The HUD is a non-issue here. There is nothing about the game's 3D balance around objects that should have anything to do with the HUD. The only exception would be if the HUD was interfering with the game's scenery somehow, and that's not the case here. Really, the HUD should be judged as a separate element (otherwise, what's the point of rendering in 3D at all? Why don't we just double up the images and judge distance by where the HUD is?).

You are correct. When we are overlapping the images this way, we are adjusting the convergence as shown here:

Image

In practice, when adjusting or overlapping a 3D scene, there should be give and take - like a seesaw. Unless the seen is so far in the distance, it should be impossible to be able to overlap almost the entire left and right image over each other. Look at the first image in this thread. Does it really make sense for the entire road and the buildings in the distance to be almost 100% identical?

Try the DDD images in this thread. Do they not look more visually interesting? Ignore the HUD and try to overlap them...can you do it?

Regards,
Neil


Tue Jan 22, 2013 11:10 pm
Profile WWW
Two Eyed Hopeful

Joined: Thu Dec 27, 2012 4:22 am
Posts: 98
Reply with quote
Hi Neil,

I apologize for my earlier statement, you did state that you thought something was missing from the driving sequences. I also should not be putting words in Andy's mouth, but my reading was that he loved the game ("One of the best looking games in 3D IMO"), and only later seemed convinced there was a flaw.

I hate to be thick, but I'm really not seeing a problem with the images in this thread. To answer your specific question, I don't really see much difference between the DDD image and the bike image, after aligned. They seem to align roughly evenly too me, although the blur effect in the DDD image makes it harder to tell.

If I may, Let's back up and talk about a specific example.

This image you rate as expected, with normal separation. Aligned image.



This image you rate as abnormal, with too little separation. Aligned image.



Now when I look at both those images using Anaglyph, I see the motorcycle is a little deeper into the scene than the character in the first image. I'm not sure that's significant, but I'm not sure it's not either.

When I look at the first image, and the lady on the left of Wei Shan, then look at the parked car on the right in the second image. Here again, pretty roughly same depth into scene. But, the separation I see there is also roughly the same. Look at the character's leg. Look at the rims of the car. The red and blue images seem to me to be at roughly the same separation.

Now look into the very deepest part of both scenes. In the first image, I can see two sort of octagonal blue and red images, to the right of Wei Shan's shoulder. In the second image, I can see two fuzzy red and blue squares above the biker's helmet. If I use my cursor as a rough measuring tool, those images are displaced at about the same amount.

Now maybe this is what you mean, because in the biker picture, those squares are much farther away and should be separated further than the octagonal images. However, I really think that is a function of convergence being dialed onto the bike, which is further into the scene. If we don't go that far into the scene, those far squares would be further apart.

Please take a look, and let me know what you think.


Wed Jan 23, 2013 3:11 am
Profile
Two Eyed Hopeful

Joined: Thu Dec 27, 2012 4:22 am
Posts: 98
Reply with quote
Also, let's take another comparison of that bike image, with a comparison to the DDD image that you prefer. The DDD car image is a better comparison case, because it goes into the far distance, just like the bike image.

One thing I'm assuming that I should make sure we agree upon- that separation is directly correlated to depth. The wider two anaglyph images are apart, the greater the depth we perceive when viewed stereoscopically.

If we agree that is true, then I've modified the bike picture to not have as much convergence, and specifically set it to try to match the DDD car picture convergence, at maximum depth.

As noted, this modification damages the HUD, but I agree with your earlier comment that it doesn't matter in these modification test cases. That separation in the original Andysonofbob picture is really extreme. I'm very comfortable with viewing S3D and that one gives me eyestrain on my projector.


Here is the bike pic, modified to make the red/blue squares above Wei Shans' helmet be a cursor 'hands width' apart. Those are also the windows on the farthest building.




Here is the DDD car pic, with no modification or alignment, as a comparison. The tall building in the far distance, directly down the center stripe is my comparison for 'hands width' apart.




When I compare those images, I see essentially the same separation of left to right, and when I look at them in true stereoscopic, I think I also see roughly the same depth.

Please let me know if you think I'm making a mistake here.


Thu Jan 24, 2013 12:13 am
Profile
3D Angel Eyes (Moderator)
User avatar

Joined: Wed Dec 31, 1969 6:00 pm
Posts: 5723
Reply with quote
Hi Bo3bber,

Sorry for the late response. My first point is that Nvidia, AMD, and DDD all have stereoscopic 3D images. There are two camera views happening here, and that was never in doubt.

As stated in the review, the issue is that the balance from one seen to the next is way off with the Nvidia & HD3D implementation. If all the scenes were handled the way the driving scenes are handled, there wouldn't be a problem because we could adjust the separation and convergence to a mix that works well for all scenes and the game would look very different indeed - even the driving scenes.

Unfortunately, what IS happening is you can get a proper balance for about half the scenes, but the remaining activities have this super weak 3D experience with ineffective camera/convergence placement. If we got the driving scenes to where we needed them, the game would be unplayable because the remaining scenes would be unviewable (divergence, etc.). Showing the difference between the left/right views is our way of exemplifying this weakness in this remaining 3D experience.

Now if we look at the DDD result, you get a great balance in all scenes. You don't have a weak 3D experience in 50% of the scenes because the convergence and separation settings hold true throughout the game. The convergence and separation behaves the way we expect them to in almost every scene.

I will try to get my hands on some additional DDD pictures to exemplify the difference. Just to add, while I don't know if this is the exact problem that Nvidia was talking about, during our review process, Nvidia expressed dissatisfaction with how convergence was handled in the game. I don't think I'm alone with this.

Regards,
Neil


Sat Jan 26, 2013 10:46 am
Profile WWW
Two Eyed Hopeful

Joined: Thu Dec 27, 2012 4:22 am
Posts: 98
Reply with quote
Hi Neil,

Thanks for the response.

I understand now. It's not that any given image is bad, it's that there is no set-and-forget convergence+depth that works for the entire game.

You might want to edit the review slightly to call that out more explicitly.


Based on the strength of the pictures I've seen here, and this awesome discussion, I went ahead and bought the game today.

On sale at Steam for $10 today. Impossibly great.


I only had the images earlier, but I'll take a look directly at the game now. I still suspect that this is a fundamental problem of the parallax budget, but will get a better idea first hand.


Sun Jan 27, 2013 7:29 pm
Profile
Two Eyed Hopeful

Joined: Thu Dec 27, 2012 4:22 am
Posts: 98
Reply with quote
Hi Neil,

I wanted to circle back around to that overlay test, where Sleeping Dogs did not seem to have much separation when aligned.

I really respect your goal of trying to make reviews objective, instead of subjective, and I think you've gone a long ways that direction.

The aligned image test worries me though, because it winds up being more subjective than I'd like. As in "nearly identical". It would be better to have something quantifiable. Would you agree?


To that end, I'd like to suggest and alternate measuring approach that would be fully objective.

How about if the test measures the actual separation itself?

If we see an image doubled up on screen, we can see the left and right images at once. We can then see the actual separation, and as we know, the wider the separation, the deeper it will appear in stereoscopic.

I'm not sure what the right metric for measuring would be, maybe pixels apart? Maybe a ratio of max to min? Not sure, just thinking out loud here.

If we had such a metric, we could then say with authority and clarity that Sleeping Dogs was weak during driving scenes. And directly compare them to indoors scenes.


I see a lot of comments about the stereo effect being too weak. In lots of forums and lots of games. I don't have that problem, and I'm not sure why. I like it really cranked up. If we could make a metric that was independent of screen resolution and physical size, we could compare the claims more directly.

What do you think?


Sat Feb 02, 2013 5:45 am
Profile
Display posts from previous:  Sort by  
Reply to topic   [ 32 posts ] 

Who is online

Users browsing this forum: No registered users and 0 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Jump to:  
Powered by phpBB® Forum Software © phpBB Group
Designed by STSoftware.