 |
|
|
|
|
|
 |
| ||||||||
 |
 
|
|
|
|
|
|
|
|
|
| ||||||||
|
|
| 
| User list | You are not logged in.
My feature request is for exporting "tile worlds" for virtual environment compositing and image based lighting.
We assume some things that a well shot pano should have anyway: The camera is nodal, and the focal length is fixed.
What I would like is the ability to use APP to determine the FOV, pan/tilt of the camera and the lens distortion. That's all. The only image processing APP would be used for is removing the lens distortion and maybe a bit of blending. The main output would be the undistorted images and a set of cards in 3D space that match the view-plane / frustum of the taking camera. They could be in a basic single polygon objects in a format like OBJ with UV maps normalized from 0-1 over their face and the name of the image corresponding to it's position in their material name.
Assuming a fixed pattern of images is shot, the "pan&tilt" data could be used by APP to guess where an impossible to match tile (like blue sky) MUST go (we are talking about stuff shot with a Pappy Wizard or Gigapan, or meticulously by hand) So like a form of fuzzy logic that uses both the feature based matching and a "projection" of the image plane by "pan and tilt). The option could be called "Guess position of un-matchable images by shot order"... ok, so that doesn't exactly flow off the tongue, but at least it's descriptive!
Perhaps "Smart Blend" could also output alpha channels for each card, or the cards could optionally be processed thought the Smart Blend engine to clean them up a bit. But the main idea here is to end up with a bunch of cards in 3D space.
How about the ability to save the lens profile and use it to un-distort video?
The 3D cards feature wouldn't just be useful to FX artists like myself. Now that Photoshop does painting on 3D, this would be a great way for people to touch up their panoramas in a more intuitive format.
My last feature request is a "pre-process" mode that could be used to merge HDR images BEFORE feeding them to the stitcher. Perhaps this mode could even be user scrip-table so users could plug-in any HDR merger or tone mapper thingy they want to use.
Offline
I´m not sure to understand you precisely - so let me pick the first point: IBL.
For IBL you need a sphere in "real" HDR.
The way to generate it is:
1) shoot bracketed images using a precise Nodal-head.
2) stitch all shots while telling the editor to sort them as bracketed layers.
2a) don´t use color-correction - it might treat the layers incoherently.
3) render them as 3 bracketed layers using the command %L behind the pano´s name in the rendering dialogue.
4) APP now will render 3 (or more) layers each representing a bracketing-stage of your shots.
5) the 3 (or more) rendered equirectangular pictures give to Photomatix to generate ONE HDR from them and save it as .exr or .hdr
5a) don´t use any kind of correction here like anti-ghosting or so.
6) import the equirectangular in Cinema4D, Maya or similar apps for being used as IBL.
Using a motorized and programmable head like the Merlin with Papywizard or with the TC-controller you can have a xml file generated to tell
APP/APG the pattern of your shots - maybe you can use the xml also to tell the 3D app´s camera where it should look to . . . . you´ll have to work that out.
The rest of what you wrote i´m not sure to understand the meanings precisely, sorry. What do you mean with "cards"? Do you mean "cube-faces"?
best, Klaus
Last edited by klausesser (2010-01-25 12:48:51)
Offline
klausesser wrote:
The rest of what you wrote i´m not sure to understand the meanings precisely, sorry. What do you mean with "cards"? Do you mean "cube-faces"?
You are correct. One can simply use an HDR equirectangular or cube projection for IBL. And for IBL, it doesn't even need to be very high res, only "accurate" in terms of the light values it contains. (For stuff like reflections, etc., it's a different story) But my real feature request is actually the part you don't understand.
"Pan and Tile" is a pretty common technique in visual effects for creating backgrounds. It's related to panoramas in that it requires a nodal pan and tilt to gather the images. However, rather than warping and stitching the images to a single giant projection, they are placed on individual cards and aligned to match the taking cameras original projection.
Here is a video that illustrate "pan and tile" / tile worlds. A "tile" is simply one image of a "panorama in 3D":
http://www.digitalartform.com/archives/ … nodal.html
Having the individual images as tiles is convenient as the renderer does not need to load the images if they aren't in the view. Video files can alternately be mapped to the tiles, creating a moving panorama (or at the very least, a panorama with windows of moving stuff in it) Depending on the renderer used, individual tiles can be very high resolution but memory usage at render time can stay sane because only the tiles in view will load their images.
So, what I want is for APP to be a helper tool in creating tile worlds by aligning the tiles, unwarping the lens distortions and exporting 3D geometry I can load right up into Lightwave, Nuke, Digital Fusion, Maya, C4D. whatever... and map those un-lens-warped images to.
Offline
lordtangent wrote:
Having the individual images as tiles is convenient as the renderer does not need to load the images if they aren't in the view. Video files can alternately be mapped to the tiles, creating a moving panorama (or at the very least, a panorama with windows of moving stuff in it) Depending on the renderer used, individual tiles can be very high resolution but memory usage at render time can stay sane because only the tiles in view will load their images.
So, what I want is for APP to be a helper tool in creating tile worlds by aligning the tiles, unwarping the lens distortions and exporting 3D geometry I can load right up into Lightwave, Nuke, Digital Fusion, Maya, C4D. whatever... and map those un-lens-warped images to.
Since i work with Maya and 4D for scenes in car-advertising i think i know what you mean. But we use high-res spheres instead of tiles in short moving sequences - despite of very hig-hres backplates (which are shot using digital MF or LF)
Using spheres the possible camera-movement is less limited: http://www.klausesser.de/Anim2.html
best, Klaus
Offline
Nice stuff!
But as I mentioned, for mapping video to tiles, a single sphere has it's limitations also. Of course, it would be possible to stitch the bulk of the sphere and line up moving tiles by eye. But what if all the tiles have video on them?
Offline
lordtangent wrote:
Nice stuff!
But as I mentioned, for mapping video to tiles, a single sphere has it's limitations also. Of course, it would be possible to stitch the bulk of the sphere and line up moving tiles by eye. But what if all the tiles have video on them?
Correct. 
best, Klaus
Offline