As I will post some renders now and again, here's where I'll start. I'll be making a mural display of 45 x 3 meters, depicting (slightly hilly, luckily) Dutch landscapes from 250.000 years ago until medieval times. But... as a continuous landscape, where people can walk by and see the terrain change.
Several problems (challenges) emerge; as it will be a huge file, making it in one go isn't feasible. So, 5-10 or so different landscapes? How to merge them? By making a long panorama, turning the camera, was my first idea, changing the terrain and growth from left to right (ice ages coming and going, veggies emerging and being cut down again when people start to populate the land....). The foreground is the most difficult to merge, as it's full of detail. Shooting images in TG in longish images but not the total panorama. Not crops but different files, or the file would be enormous, I'm afraid.
So, as a starting point; if anyone has any ideas or experience in this kind of venture, I'd welcome any advice!
So you got this crazy commission eventually, congrats Ulco :)
At first I am thinking of making one base terrain, which you will use for the entire panorama.
Then, depending on the number of "time-zones" or "eras" you need to depict, set-up cameras which will split the panorama in appropriate pieces.
Use software like PT stitcher. BigBen had some interesting posts recently regarding stitching and fixing lens distortions:
http://www.planetside.co.uk/forums/index.php?topic=15047.msg146633#msg146633
This would allow you to calculate how much overlap you would need for each piece.
As example: if one of your tile has a forest and in the next tile you want to depict the start of wood-cutting then make sure the forest is in the seam.
Wood cutting would be depicted at the left in your next tile, using the transition from the previous tile. If you see what I mean.
As long as you can keep those elements flowing you may be able to get away with splitting this project in X number of pieces.
Good luck and let me know if you need any ideas and/or advice!
Cheers,
Martin
Thanks Martin. One terrain is for sure. The most important problem is the camera placement. I think it should be a circular panorama, basically shot from one point. If you set up parallel camera's the foreground will never match (take 2 trees/stones behind each other, they will be viewed from different angles and can't be stitched). But if you shoot from one point it is hard to get differences in the foreground (ice age, no ice age). The guy who wants this is aiming at eye level POV's, no birdseye shots.
I think I must advise him to raise the camera at least as high as tree level, to be able to more easily stitch/blend foregrounds. With a bit (much) fiddling in Photoshop that is probably the most effective way.
Experimenting is paramount, so I'll get to it.
Anyone else?
There are certainly a number of ways this can be handled. Do you have a sense of how they want the end result to look in terms of the blending? Is it intended to look like one continuous view with "natural" blending between eras, i.e. no image blends/fades? The latter would certainly be much easier, just setup different versions of the same scene, render out the same view of each in a wide format, and blend each time period in as desired. Of course there would be a lot of render overlap, and the look of it would be more simplistic than if you tried to somehow integrate the time periods into the actual construction of a single scene. For that I might suggest using camera-projected masks to control blending of each time period, just as a base concept. Perhaps better yet, use 1 camera as a base mask for each time period, but have a wide aspect ratio render camera that *does* actually render the whole scene in one go (rendered in tiles on a render farm, of course). Then you potentially avoid some of the distortion issues of stitching and rotation-based panoramas.
Out of curiosity, what resolution do you need to deliver the final image file(s) at? Or, put another way, what resolution do you anticipate the final overall image(s) for printing to be?
- Oshyan
It should be one continuous landscape. I think I get what you mean; shoot the same terrain (but different details/eras) in one very wide angle (say 3000 high by 45000 wide), but use crops for each era and blend these together. Ah! Good thinking.
Also the camera mask is a good idea, you got my brain working again. Thanks!
Resolution... for a wall of 3m high @ 20-40 dpi (billboard resolution) it should be 2400-4800 px high, and very(!) wide. So it's going to be a farm, point.
Added 2 tests
Yes, you've got the basic idea I was talking about. :)
These early tests already look quite promising. The sense of "heather" or small ground cover in the 2nd image just from the terrain shapes/displacement and coloration is quite good actually.
- Oshyan
Quote from: Dune on November 26, 2012, 02:37:11 AM
Thanks Martin. One terrain is for sure. The most important problem is the camera placement. I think it should be a circular panorama, basically shot from one point. If you set up parallel camera's the foreground will never match (take 2 trees/stones behind each other, they will be viewed from different angles and can't be stitched). But if you shoot from one point it is hard to get differences in the foreground (ice age, no ice age). The guy who wants this is aiming at eye level POV's, no birdseye shots.
I think I must advise him to raise the camera at least as high as tree level, to be able to more easily stitch/blend foregrounds. With a bit (much) fiddling in Photoshop that is probably the most effective way.
Experimenting is paramount, so I'll get to it.
Anyone else?
My idea is almost exactly the same as Oshyan's only, as usual, he's better in explaining it. Perhaps I should PM or reply you in Dutch here.
The use of 1 render camera and crops for every timezone/era might work better or more easy.
It would avoid lens issues you mentioned, although those can be overcome using the tools I referred and linked to, plus that you can't get a circular like image if you use 1 single camera.
You'd need multiple cameras for that and additional overlap.
Like I said the seam-issues can be fixed using the proper tools.
For every crop you would need to make a separate tgd anyway, because likely you'll use quite some models and props and memory consumption would go through the roof. Your medieval city already took quite some RAM only for loading the project, let alone rendering it :)
Indeed, I didn't extract the idea from your explanation, Martin. Sorry about that.
I will save all different 'crops' as different tgd's, and use the camera mask for populating, pointing it differently in every tgd. That should keep the populations small enough. The positive from a low POV is that I don't need huge populations. A medieval village won't be visible much either, distant houses being obscured by the front line. Much easier than a high POV.
The starting point will now be a 45000x3000 terrain view. If I get the hills alright, then it's a matter of ground cover.....
Sounds like a great project in more than one sense!
If you do it by turning your camera instead of moving it in a linear way, and if the result is supposed to be 15 times wider than it is high, I think you will need more than 360°?
I didn´t calculate any camera settings, thats just my rough estimate.
That would mean that your oldest szene and your newest one could occupy the same space, which would be cool in my opinion. As if you turned around veeeery slowly while ice ages passed, looking at the same but changed spot again in the end.
Of course I don´t know if that´s the intended effect, because the people will in fact be walking by instead of turning on the spot.
I am courious to see where this goes!
Best Regards,
J
Interesting project,indeed.
Will the mural be on a straight wall or on a curved one (like some pano-paintings)?
Thinking of vanishing point(s) problems here when you actually walk by.
If your client wants a close up foreground there could easily be some serious
distortions and you would have to use some optical trickery like depth levels or so.
Just thinking,anyway,
good luck,J.
I don't know yet what kind of wall it'll be. With one large file there's no problem with vanishing points or 360 degrees or more.... Just came home and made a quick setup with 3 cams. Only the left part (displacement intersection, blended by a camera mask) doesn't work properly yet.
The main problem now is that it's hard to work in a window that is so restricted. I suppose there's no way to zoom in on the 3D preview to fill one part from top to bottom?
Hmmmmm... seems to be a lot of stretching on the left side of the image,
perhaps try a setups with Orthoginal cameras ?
The stretching is something else, probably a camera (mask) where I forgot to change the position to the new render cam position. In this previous test it isn't stretched. Orthogonal cams came up, indeed, but I haven't tried it yet. It would be so much easier if I had a nicer ratio in the 3D preview to work in. Perhaps some temporary camera and then another temporary render ratio?
I would suggest having 1 camera for the "full scene view", 1 for each "era", and then one with the same base position that you can pan left and right (keep the same vertical FoV as the final render), using the camera pan tools. This way you can view any part of the scene and render tests while maintaining at least *some* sense of the bounds of your scene (vertically if not horizontally - you could place reference objects at the extreme left and right of your view for the horizontal).
- Oshyan
Hi Ulco. congratulations on getting a new commission, especially since its in a subject that interests you! Its good to get paid to do what you like.
I agree with lots of whats already been written, but I wantted to chime in in my own words:)
Lets assume you have all the resources you could ever need. HA HA, I know, there are always restraints, but just as a starting point...
For me, the *ideal* way to go about this for highest quality results would be to do A full panorama at full size and resolution for each period in time of the landscape. So each historical age, from first to last, of the same landscape as it changes over time.
After getting (4-6-8?) panoramas, then I would take them into photoshop as layers and paint a transparency mask so that the oldest landscape started on the right and ended with the oldest on the left. (I think this is what you were saying) This would give you the best control over where and when one age transitions into another.
I asume erosion and geological transformation will be extreme, i.e. Ice age, continental displacement, (volcanic activity?) Forest growth.
In my photography I always shoot vertical when doing a pano. with a photo camera its very important to remember to make sure the cameras pivot point never moves despite changing views between frames. The same should hold true for a virtual camera in terragen2. Shouldn't it?Anyway, doing so will make stitching much cleaner. I also usually overlap my frames by 1/3rd to 1/2. Its not usually but ensures good coverage. I doubt you'll want need to do that with a render. But Ideally, it would be nice to have that much material to work with when stitching and painting, masking.
I usually shoot HDR panos, so my workflow may be totally different than what your looking for.
For panoramas I use autopano pro http://www.kolor.com/ I have tested it on a TG pano. I did not run into anything unusual compared to working with photos, that I can recall. It was a while ago.
Just what came to my mind. Its a very interesting and complex technical project. I hope their giving you what you need to do what they want
Badger, I briefly mentioned your approach earlier, although perhaps not as clearly.
QuoteThe latter would certainly be much easier, just setup different versions of the same scene, render out the same view of each in a wide format, and blend each time period in as desired. Of course there would be a lot of render overlap, and the look of it would be more simplistic than if you tried to somehow integrate the time periods into the actual construction of a single scene.
What I'm interested in is if this project is intended to have a more literal "integration" of the different time periods across the image space. In other words, rather than being a simple fade, or any other kind of work you could easily do on an image masking basis, to instead have the elements of different epochs *literally* blending into each other left-to-right. To give you a simplistic example, let's say there is ice age on the left and modern times on the right. In the middle is the blend. On the left you create large glacial terrain shapes and color them white and blue. You have a camera-projected mask that causes this texturing to fall off around the middle, blending into green texturing on the right. You have trees covered in snow on the left, in the middle they are intermixed with barren trees with no snow, which gives way to fully leafed treas on the right. These are not simplistic fading image blends, but unique instances of objects with these characteristics. Does that makes sense?
Now I am *not* sure if such an approach would be notably superior to a more simple image blending technique, in fact it might end up looking worse. But from a technical standpoint it's far more interesting, and I think that's what Ulco is attempting here. So I'm quite curious to see how it goes!
- Oshyan
Oh hey Oshyan, sorry, I was editing. Yes I saw your post and agreed with it.
QuoteWhat I'm interested in is if this project is intended to have a more literal "integration" of the different time periods across the image space. In other words, rather than being a simple fade, or any other kind of work you could easily do on an image masking basis, to instead have the elements of different epochs *literally* blending into each other left-to-right. To give you a simplistic example, let's say there is ice age on the left and modern times on the right. In the middle is the blend. On the left you create large glacial terrain shapes and color them white and blue. You have a camera-projected mask that causes this texturing to fall off around the middle, blending into green texturing on the right. You have trees covered in snow on the left, in the middle they are intermixed with barren trees with no snow, which gives way to fully leafed treas on the right. These are not simplistic fading image blends, but unique instances of objects with these characteristics. Does that makes sense?
Now I am *not* sure if such an approach would be notably superior to a more simple image blending technique, in fact it might end up looking worse. But from a technical standpoint it's far more interesting, and I think that's what Ulco is attempting here. So I'm quite curious to see how it goes!
Oh yeah! Now were getting to the meat of it!:)
For me, I am assuming that Ulco will give a ton of attention to the facts of the time periods, just because his historical renders always seem to me to have that information. Of course I am NO expert ecological/geological history, and less so when it comes to Ulco's part of the world. But the renders appear to have all the literal facts of nature in them.
So if its then just a question of masking later, or building the information into as few renders as possible, I have to support the masking. THere will be nothing simple about it. I am thinking in terms of matte painting now. And final output. So while I must agree that "
from a technical standpoint it's far more interesting" because it is. I think that will require a ton of complex experiments with-in TG2. While masking is a proven method for panorama making. Where also the final esthetic can be more easily pre-visualised, since there is already a ton of work out-there to reference.
Maybe I am mistaken now, but it sounds like the choice is between building a world with all of the elements in one place (sort of), and painting the elements into one place from several TG2 worlds (sort of).
building it would be, like your suggesting, AMAZING. I often sit here with my jaw dropped open in awe of the technical genius of many of you guys. But I would be terrified to attempt something so complex *after* the Commission began. I would want to know how thats going to work from the start.
On masking the transitions, I am not suggesting anything simple. Definitely not some kind of gradient of one age to the next. But rather a painting in of the visual facts at the meeting points of the time periods.
Imagine the space on a print where one period of time meets up with the next. What is happing at that point? Its it a hard transition? Or is there a representation of the period of time between the two major ages? But going the mask and paint method there is a lot more options. All of which can be more easily changed and modified in photoshop.
I guess now after reading and writing here with you. The *ideal* would be to do both, and then have the freedom to mix an match.
But like I said I would be terrified to try the more complex technical approach without first having done successfully before. Then again, I don't deny for even a second that Ulco's and your (Oshyan) technical knowledge of TG2 and rendering is vastly above mine. I am just interested in the ideas here. Its very cool that your getting paid to do this stuff Ulco, and everyone else who earns while having fun!;)
*edit
Oops sorry, when I say painting, I am thinking in "matte painting" terms. Just to be clear. ;D
Understood, so you're thinking of a reasonably sophisticated approach as well, just using post image editing rather than in-scene (and, yes, somewhat more "rigid"/inflexible) construction. That being said there are definite disadvantages to the post blending approach, not least of which is likely longer render time (since you have to render at least *somewhat* larger than each time period's area of depiction will display as in the final, to handle overlap and creativity in post), as well as huge memory requirements for handling multiple layers of imagery at that high resolution in an image editor.
The nice thing is Ulco already has a proof of concept going for the "do it all in one scene" approach. Granted it's a long way from full realization at a high level of quality, but it proves the basic idea is sound, and he put it together extremely quickly.
I think no matter what approach is taken, it's going to be complex, challenging, and resource-demanding.
- Oshyan
Ahh yes, I did forget that Ulco is talking about printing in "Meters". That would be a heavy lift in photoshop. That fact is rather important, now that you brought it back to my mind. Really I have no idea what would happen then. I have never worked that big. The largest .psd I have dealt with is 1GB. If I remember right I was usingCS3 then. It was very cumbersome. Although Aoutopano giga is desined for this kind of lifting. I posted this link before, but in light of this conversation it may be more relevent now. http://www.paris-26-gigapixels.com/index-en.html details: step1 http://blog.paris-26-gigapixels.com/en/?p=115 , step 2 http://blog.paris-26-gigapixels.com/en/?p=110 , step 3 http://blog.paris-26-gigapixels.com/en/?p=114
Anyway, whats really important to me is that I don't want to wait any longer. I want to see the finished version of this thing now!
Work faster Ulco! I need my super complex terrain visualization fix now ;D
QuoteBut I would be terrified to attempt something so complex *after* the Commission began. I would want to know how thats going to work from the start.
So am I, maybe not terrified, but a little anxious if (mainly) the system can handle it. It will be huge in terms of kB's... Mb's and maybe GB's. You're right about the approach I want to take, make it one large image, with no blending afterwards, but probably rendered out in crops. It will have to be printed in strips anyway, the largest they get here is 2 meters wide.
I was thinking along your camera line as well, Oshyan, but being very busy at the farm (builders work, never done it before) I only have an hour or so per day to spend on this, for the time being. If I can restrict trees to the cropped/masked areas for every era, and save as separate tgd's, I hope to keep the count low. I will also probably work with fill lights (no GI) or I have to do every large GI prepass myself and upload each. Unless the farm can handle that properly by then (Ty?).
There won't be very much erosion and lava stuff and such, as it's a Dutch landscape, with low rolling hills at most. Such as in my first test. Pff, that's a relieve...
It's not a matter if the renderfarm can manage that properly, because TG can't yet at the moment.
So you probably need to generate the huge cache yourself, which made me think:
What if you save "crop GI caches" of the entire frame and use the blending for animation option in the GI settings for blending your GI crops when rendering the final frame? Would the black parts, outside the crop, be replaced with GI from other cache files then?
I tried a skybox once and set up an animation for my camera for each tile of the box.
I forgot to switch off motion blur and each tile had some ghosting of other tiles and in tile details.
I think I should try this, low scale.
Well, instead of venturing into that and having to upload an enormous GI cache file, I'll stick to fill lights. I think I can make the lighting for this purpose good enough, with a little effort. But thanks for co-thinking, Martin!
It's too bad I have so little time at the moment ( I was working on drainage today, hard labor!).
Quote from: Tangled-Universe on November 28, 2012, 05:10:30 AM
It's not a matter if the renderfarm can manage that properly, because TG can't yet at the moment.
So you probably need to generate the huge cache yourself, which made me think:
What if you save "crop GI caches" of the entire frame and use the blending for animation option in the GI settings for blending your GI crops when rendering the final frame? Would the black parts, outside the crop, be replaced with GI from other cache files then?
I tried a skybox once and set up an animation for my camera for each tile of the box.
I forgot to switch off motion blur and each tile had some ghosting of other tiles and in tile details.
I think I should try this, low scale.
This does work, I've tried it before. :-D It takes a bit of manual work, you have to render each cache manually and name them sequentially so it can find them all. But the end result is smooth GI.
- Oshyan
Ah nice Oshyan, good to know!
First larger rough test @ 10000px wide. I think it will be a bit strange if there are too many transitions, but we'll see what the client says.
It's woooorking! Cool, hehe.
- Oshyan
Yeah, it is, but I'm not there yet. Now I have to figure how to do the terrain; either as one wide vista, like a panorama, with maybe 5 hills along the whole 45 meters, or as a continuous sequence of vistas with say 5 hills in each era. Now an ice age will only occupy part of a hill, if you know what I'm getting at, not a total landscape of a few hills before we move on to the next ear with again 5 different hills.
I haven't conferred with my client yet, as I first want to find out for myself what's feasible and gives the best result.
I also had to use a trick to get the displacement intersection work when blended by a camera driven mask, but that seems to work. This render at 0.5 and AA4 took only 1.2 hours, so the final should be doable in crops.
Wow Ulco. Is that what the country side looks like where you live? The rolling green hills and few trees? I would love to walk out there all alone! For me that kind of place is other worldly, there is nothing like it where I am from.
I think I am seeing an issue that will stick out to me though. If you are doing this in one render then I would guess you are able to get such a wide shot by cutting out nearly all of the foreground and most of the sky. In a photographic panorama, the individual parts would have been shot in a vertical "portrait" position, think a 8w x 12h aspect ratio, whatever ratio that is, just as an example.
The shots would over lap from one side to the other (as much as you like) and the result would be that you get back the foreground and as much of the sky as you like. Though the pano narrows at the far sides of the image requiring you to crop.
Non of this maters if you are printing a bill board. But if its something anyone will be allowed to get close too. the image as is will look like a crop of the distant parts of a photograph. In a photograph the foreground is alway more crisp and clear than the background. Right?
So I am not sure how you got such a wide aspect ratio, but something to consider.
No this is supposed to be the south of the Netherlands, near Belgium and Luxemburg, where we do have some rolling hills. Check out images of 'Limburg landschap'. The rest of Holland is quite flat. This is 1 mile from where I live, a heathland (National Park) in September. But it's indeed green and not very populated where I live, luckily.
It's supposed to be about the landscape, so a sky is of minor importance, but I could always point the camera slightly up for more sky. The only thing I wondered yesterday is whether to change the camera angle from 60 degrees to (very) wide angle in order to get more hills within the width. Have to check that. I hope to get away without having to stitch parts together. Working in this file is very cumbersome >:(
QuoteCheck out images of 'Limburg landschap'
Ok thanks. These landscapes are much like where I live.
Keep going Ulco! I am very interested in the process, and what you find will work.
Update, but now stitched together from 8 (ratio 1:2) turning cam's renders. Rendering (and working) in one stretched (1:15) image ratio is very cumbersome (to say the least), and this would be much easier. Only now the stitching will take its toll from my machine, I'm afraid. So perhaps I can work in this file and then switch to a final 1:15 render.
Looking forward to see more of this.
This is the region of the Netherlands where I grew up so it's very dear to me and in my childhood it wasn't as populated as nowadays so I can somehow relate to what you're rendering.
Updated a bit.
Masterful!
Looking very promising. Are you now settled on the stitching approach for the project instead of a single scene/render?
- Oshyan
No, I still prefer a single shot (min 2400x36,000px for 20dpi, max 4800x72,000px for 40dpi). The problem with stitching is that the files will be quite big, and I don't know if my pc can cope with that. Especially since I will loose some edges here and there, so the renders should be much larger as well.
Also, with stitched images, I can only turn the camera and make a half circular panorama. So the sun will rotate with the views. That might be interesting in a circular exposition, but on a large straight wall a sun in one position might be more realistic.
Like this:
Would using a camera with a 20 degree fov work. Each section of the whole image is made by rotating the camera to the edge of the last image. 20 degrees would have little distortion.
I assume you then get a telephoto effect, bringing the distance closer in, but I can try. I did try all sorts of things, even changing the film size to 45x3, but this seems the best way. Even a little distortion will force me to stitch large files, with this setup I can just render away, be it in crops. They'd have to be printed in panels anyway.