[Nova66]

are you try multilght with panorama ? or it's impossible ?

We're getting off the topic of this post but you bring up a really good point.
As I've tried to explain in this post:
http://www.maxwellrender.com/forum/view ... 37&t=43766
Light emitters don't work properly with the Spherical & Lat-Long lenses. So the answer is that Multilight does NOT work properly with panoramas


Andrew.

[seghier]

Thanks ; maybe the developer have an idea !

[Dover Studios]

Hi Nova66 and seghier.

To make it easier to for you guys to understand how the LatLong Stereo and Fish Stereo lens shader controls work inside of the Maxwell Studio I added several new topics to the main Domemaster3D wiki. It is a bit late here so please forgive any spelling errors that might have slipped through in the first pass of composing these Maxwell based wiki topics.

I'm confident that Maxwell is going to add more stereo lens shader notes and documentation to their online knowledge base too so there will be even more information out there to help users.

Here are the current Domemaster3D Wiki Topics on Maxwell Studio:

Maxwell Studio LatLong Stereo
https://github.com/zicher3d-org/domemas ... ong-Stereo

Maxwell Studio Fisheye Stereo
https://github.com/zicher3d-org/domemas ... eye-Stereo

PanoView for Maxwell Render
https://github.com/zicher3d-org/domemas ... ell-Render

* * *

Nova66, I'm really happy to see that someone has already explored the CubeX demo scene in Maxwell Studio. Boy are you ever fast!

Once the Maxwell 3.2 stereo lens shader features are all ironed out, I'm hoping that one day I will see a panoramic version of your Bentley Microstation based "Render Ready" red and white Rocket-ship model:

http://communities.bentley.com/products ... 750#199750

Regards,

Andrew Hazelden
Domemaster3D Co-Developer

[Nova66]

I added several new topics to the main Domemaster3D wiki

The Wiki is looking pretty good Andrew. There are two minor things that I noticed that may or may not be relevant...

1) Your documentation talks about the Screen Space Texture Maps being located in:
C:\Program Files\Domemaster3D\sourceimages\
It might be because we are getting your stereo lens shader through Maxwell but I don't have that folder on my system.

2) Your screen capture of the texture picker has got some strange numerical values in it.
It might be because you haven't loaded an actual image but I would have expected normal numerical values to look more like this:

I'm hoping that one day I will see a panoramic version of your Bentley MicroStation based "Render Ready" red and white Rocket-ship model

Wow, I'm impressed that you found that
I have to give credit and point out that the red and white Rocket-ship is based on Tintin's Destination Moon cover page:


I'll have a go at doing a Stereo panorama of the Rocket-ship scene but the model is pretty simple and doesn't have much to fill out a full 360° view. I may have to re-work the model to make it more visually appealing at close range


Andrew.

[Dover Studios]

I loved reading the Tintin books growing up.

1) Your documentation talks about the Screen Space Texture Maps being located in:
C:\Program Files\Domemaster3D\sourceimages\
It might be because we are getting your stereo lens shader through Maxwell but I don't have that folder on my system.

2) Your screen capture of the texture picker has got some strange numerical values in it.
It might be because you haven't loaded an actual image but I would have expected normal numerical values to look more like this:

1. Yes, you would only have the Domemaster3D folder on your hard drive if you used the Domemaster3D installer for Maya/3D Max.

The control texture maps aren't that complex of a thing to customize on your own in Photoshop once you see a screenshot of what they can look like. You are simply choosing where you want to reduce the stereo effect in the rendered imagery and painting that area black in the texture map with a soft brush to cause a smooth fall off which helps blend the transition zone. Also, since the control texture maps are stretched and squashed using "screen space" coordinates to line up with the rendered image in the render view, you don't have to worry about them matching the exact render resolution or aspect ratio of the final image.

The same images can be downloaded on the project's GitHub page here:
https://github.com/zicher3d-org/domemas ... er/maxwell

2. The strange numerical values were from not having an image linked into the dialog at that point in time like you suspected. I've updated the wiki page with images of a LatLong and fisheye separation maps loaded in the texture picker dialog.

Regards,
Andrew Hazelden
Domemaster3D Co-Developer

[Dover Studios]

Hi Nova66. I noticed the following stereo related post of yours in the 3.2 thread from a little while back:

Found some news... http://support.nextlimit.com/display/mx ... era+Lenses

That's a great find
The Lat-Long Stereo Lens sounds really promising but I'm not sure why it generates three images (Left, Right, & Centered). I'm wondering what the purpose of the Centered image is and will it take three times as long to render instead of just two times for the usual Left & Right eye images?

The reason for the center camera option is that it provides the ability to render a monoscopic "2D" style LatLong output without any of the stereo adjustments or control maps effecting the footage.

You would typically render just the center view, or a set of left and right stereo pair images. Not all three image outputs at once.

This center mode is useful if you want a simple way to render the scene with the Lat-Long Stereo shader applied and you haven't decided on the zero parallax distance adjustments or camera separation results you want in the final imagery.

Regards,
Andrew Hazelden

[Nova66]

The Lat-Long Stereo Lens sounds really promising but I'm not sure why it generates three images (Left, Right, & Centered).
I'm wondering what the purpose of the Centered image is and will it take three times as long to render

The reason for the center camera option is that it provides the ability to render a monoscopic "2D" style LatLong output

I realise that now and the third centre option is a welcome addition. I wrote that initial comment before I'd seen for my self how the new stereo lens shader actually worked. My mistaken thought was that the stereo lens shader would output both the left & right eye images at the same time and it seemed wasteful for it to be calculating a third central image as well when it may not be needed.

All is good now that I see how it works
Andrew.

[Nova66]

Hi Andrew Hazelden & Juan,

The problem with the warping at the poles of the Lat-Long Stereo Lens is still present in Maxwell v3.2.0.2, I can't see that there has been any improvement at all
I feel like the Separation Map is the key to the problem but it makes absolutely no difference to the output weather I use a Separation Map or not.

Could you guys go back to my very first post in this thread and download the sample scene that I provided and post the image for the right eye render so I can confirm that it's not just me that can't make this thing work properly?


Thanks,
Andrew Novinc.

[Dover Studios]

Hi Nova66.

I am happy to mention that Juan and the team added Maxwell Studio procedural support for the stereo lens shaders which means it is possible to use Gradient3 or other texture generators to create the latlong stereo separation map on the fly.



Also the Parallax Distance control had a scene unit measurement issue fixed so it works in cm scale just like the camera separation units do.

I haven't had a moment to check out the results from the latest Maxwell 3.2.0.2 build but will take a look at your test scene this weekend and post some rendering comparisons with and without the screen space separation map.

Cheers,
Andrew Hazelden

P.S. The Domemaster3D lens shaders use a converged panoramic stereo rendering style, not a parallel stereo rendering mode.

Could you guys go back to my very first post in this thread and download the sample scene that I provided and post the image for the right eye render so I can confirm that it's not just me that can't make this thing work properly?

[Nova66]

procedural support for the stereo lens shaders which means it is possible to use Gradient3 to create the latlong stereo separation map

Hi Dover Studios, this will be very convenient. I wonder if this procedural texture can be loaded into the Separation Map slot automatically since it is always going to be needed for the Left & Right eye images.

I haven't had a moment to check out the results from the latest Maxwell 3.2.0.2 build but will take a look at your test scene this weekend and post some rendering comparisons with and without the screen space separation map.

Have you had a chance to try out the test scene. The main point of that test scene is to put geometry at the two poles so we can easily see weather or not Maxwell is reducing the parallax to zero as your gaze moves closer to either of the two poles. The way it stands now, Maxwell is not reducing the parallax at all and you get an impossible to look at stereo panorama that warps at the poles in different directions for each of your two eyes

P.S. The Domemaster3D lens shaders use a converged panoramic stereo rendering style, not a parallel stereo rendering mode.

I assumed this is how it worked and that's what the Parallax Distance was for. Why do you mention this, is there something that I'm misinterpreting?


Cheers,
Andrew Novinc.

[Dover Studios]

Hi Nova66.

Have you had a chance to try out the test scene. The main point of that test scene is to put geometry at the two poles so we can easily see weather or not Maxwell is reducing the parallax to zero as your gaze moves closer to either of the two poles. The way it stands now, Maxwell is not reducing the parallax at all and you get an impossible to look at stereo panorama that warps at the poles in different directions for each of your two eyes

It took me a few extra days to get the full range of stereo tests done with your original "Lat-Long_Stereo_Bug.zip" test scene as the basis of the stereo image analysis.

I've assembled this package of resources that includes all of my Maxwell based comparisons on the separation map settings, and a new Maya / Mental Ray Domemaster3D version of your test scene for reference (84MB):

http://www.andrewhazelden.com/projects/ ... _scene.rar

I compared the separation map attributes and can confirm your suspicions that the current Maxwell Studio 3.2.0.2 build has no difference whether there is no separation map applied, an image based separation map, or a procedural gradient4 map is applied. On the slight upside of improvements in the latest update, at least Maxwell doesn't crash any more when a procedural gradient is applied.

As far as the camera separation and parallax distance goes, I noticed that the latest Maxwell Studio 3.2.0.2 build updated the Parallax Distance attribute text to have the unit descriptor of "cm" written next to it in the GUI but the internal code appears to be still in "mm". That took me a while to work out as I was doing round robin testing of the separation map settings at the same time. hehe.

Since the outer ring of blue sphere balls is 2.2 meters from the origin I set the parallax distance to 2200 mm. Then for a comfortable stereo setting I used a mild 1/120th ratio of the parallax distance to come up with a camera separation value of 1.83 cm.



* * *

Since no comparison would be complete without brining more tools to play, I exported your scene from Maxwell Studio and re-built it in Maya 2016 using instances for the spheres and native Maya surface materials and physical sun lighting so I could do a direct comparison of the same settings in the mental ray Domemaster3D version of the latlong stereo lens shader with the exact same stereo control texture map used as the Maxwell test scene.



* * *

Here are the comparison images for your test scene (parallax distance 2200 mm / camera separation 1.83 cm):

Maya 2016 / Mental Ray / Domemaster3D v1.9.1 with a separation map applied which is the correct effect desired:


Maya 2016 / Mental Ray / Domemaster3D v1.9.1 with an empty (all while) separation map applied:


Maxwell 3.2.0.2 with texture based a separation map applied (no effect on stereo in the poles):


Maxwell 3.2.0.2 with procedural gradient based a separation map applied (no effect on stereo in the poles):


Maxwell 3.2.0.2 with no separation map applied (no effect on stereo in the poles):

P.S. The Domemaster3D lens shaders use a converged panoramic stereo rendering style, not a parallel stereo rendering mode.

I assumed this is how it worked and that's what the Parallax Distance was for. Why do you mention this, is there something that I'm misinterpreting?

That was just a note about the fact that with a converged stereo solution, having the parallax distance set drastically closer or farther away from the main object of interest in the scene will adjust the convergence and make it easier or harder for your brain to fuse the image.

Regards,
Andrew Hazelden

[Dover Studios]

I've got some really good news for you Nova66!

Juan has made a quick stereo lens shader extension compile this afternoon after seeing the latest round of test files I uploaded using your test scene.

I was able to run your scenes through this new shader build and the stereo nadir and zenith separation texture map issues, and the parallax distance cm vs mm units issues appear to have been finally solved.

From what I have heard this update will make its way into the next beta release as soon as the new installer is packaged so you will get you hands on the updated lens shader soon.

Here are some Maxwell Render vs Maya/Mental Ray/Domemaster3D V1.9.1 side by side stereo rendering comparisons with your Maxwell scene. The same stereo settings were used for the parallax distance / camera separation (1.8 cm), along with the same separation texture map.

Maya 220 cm parallax distance set to the distant blue balls:



Maxwell 220 cm parallax distance set to the distant blue balls:



Maya 80 cm parallax distance set to the close to the camera green vertical strips:



Maxwell 80 cm parallax distance set to the close to the camera green vertical strips:



Regards,
Andrew Hazelden

[Nova66]

I was able to run your scenes through this new shader build and the stereo nadir and zenith separation texture map issues, and the parallax distance cm vs mm units issues appear to have been finally solved.

That's fantastic news Andrew, and thanks to Juan for taking a look at the code. It's taken a while but we finally got there

It was great to have that direct comparison to Maya's implementation of the Lat-Long lens, there was just no denying that something was wrong with Maxwell's lens shader.

The only thing that surprised me about your notes was that you set the IPD to 1.8cm. I know you did this because of comfort reasons but in the real world my IPD is 6.8cm and I look at things closer than 2.2m all the time without any discomfort. I guess I'll just have to experiment with that setting and see what works best


Thanks,
Andrew Novinc.

[Dover Studios]

The only thing that surprised me about your notes was that you set the IPD to 1.8cm. I know you did this because of comfort reasons but in the real world my IPD is 6.8cm and I look at things closer than 2.2m all the time without any discomfort. I guess I'll just have to experiment with that setting and see what works best

It really is a subjective decision about how strong of a stereo effect you want. When I'm working on a random stereo scene from someone else that I haven't taken a long time to analyze the scale on and do several test renders for comparison, I tend to come up with a quick and generally pleasant workable camera separation setting by dividing the parallax distance with value ranging from 1/30th (for a strong stereo effect) to 1/120th (a very subtle amount of depth). You easily can go beyond that range if there aren't any elements placed in your scene that are either extremely distant or super close to the camera compared to the placement of objects near the parallax distance.

If you don't have a head mounted display like an Oculus Rift and are viewing stereo LatLong renderings using a red/cyan anaglyph formatted image that is loaded into a panorama viewer like Kolor Eyes, the anaglyph style "retinal rivalry" between the cheap left and right eye color lens filters causes more visible color fringing when you have a strong stereo effect and can be a bit distracting when you are looking at the imagery.

With a full color over/under or side-by-side style stereo image viewer, or an Oculus Rift HMD you can have stronger stereo depth effects and the brain is better able to fuse the image without the left and right image cross-talk happening like anaglyph images have.

Also, as a stereo scene development note, some people who are doing more complex stereo work will segment their 3D scene into a foreground, midground, and background set of render layers when rendering and then they can precisely sculpt the amount of depth differently for objects on those three separate layers to accentuate certain objects or reduce the attention paid to distant background elements by flattening out their depth a little.

Cheers,
Andrew Hazelden

[juan]

Hi,

In the next very few days we will upload a new beta that should fix the problems ignoring the separation map, being it a file texture or a procedural texture. Thanks a lot for your help on this!