NZ orbital WIP (& hi res tiles)

[bigben]

I haven't surfaced the North Island yet (uses a blue marble image texture), there's a major screw up with one surface and I have short list of fixes to do... but I still went wow when I saw this at full size.

[Will]

how did you get the water like that Ben? I've doing something similar (but with my own continents) but I'm trying to figure out how to get the rippl;y wave like effects on the water. great work though, keep it up

[Oshyan]

Great work so far Ben, fantastic detail and amazing scale! I'm sure you have all this in mind already, but I'll make a few critiques nonetheless.

The most obvious is the cloud scale (and possibly color). A better cloud setup overall is going to help a lot in terms of realism and in this case I'd just as soon have left the clouds out and save render time while you're still tweaking the texturing, etc.

As far as texturing itself goes, the masks look great, it seems like just some coloration for the South island needs adjustment - more browns from what I can see, on the mountain chains particularly:
http://www.4x4adventures.co.nz/p/space.jpg
http://earthobservatory.nasa.gov/Newsroom/NewImages/Images/NewZealand_TMO2002296_lrg.jpg
(I presume both of those are not too color adjusted to be valuable, and I know you already have appropriate references)

Lastly the color of some the South island interior lakes, which again I imagine you're aware of.

All in all this is coming together really nicely. I wonder at the level of input vs. output reward with all the multiple masking techniques and whatnot, but it's definitely stuff worth exploring. Keep it up!

- Oshyan

[cyphyr]

This is real. I mean really brilliant. You've sorted out the lakes and rivers nicely. I wonder how long it takes to render at these resolution. I guess its got to be at 1500 x1500 just to see the detail. What settings are you outputting at and what system are you running on (if I'm not being to impertinent :)
Richard

[nvseal]

One of the most amazing space shots I've seen.  I can't stop looking at it.

[bigben]

hmm... let's see  

nvseal and I have been exchanging material on orbital renders so lookout!

The two worst surfaces were brown (distribution completely screwed) and bare rock (colour sucks) so you managed to pick most of them    I have some potentially cool stuff coming for the lakes.

You can never have too many references. Thanks for the links

And finally... I want to render this much larger, which will show a far greater reward. On this image it is particularly interesting to see 4-5 large rivers clearly defined (they all have very wide riverbeds) based on a colour selection made on a single river (partly luck, partly true to theory)

The level of input in this case is not that great. The mask image is at the same resolution as the terrain (90m) and is under 8000 pixels. Even if I add a mask for the north island that's still fewer mask pixels than my old Grampians project which used 3 x 8000 pixels masks and 2 x 2000 pixel masks. While my new technique allows me to create even more masks, I'm struggling to use them all as it's also allowed me to simplify the masking process.  For this scene it will probably boil down to masks for grass (dead and green in a single mask), trees, bare rock, sand, snow (undecided on that one yet) and inland water (shallow and deep (hacked)).

Manually drawing rivers is pointless at this distance though and that will save me a LOT of time, so I'll probably stick with this distance to fine tune the technique and then start another project closer in.

TG seems to be handling the multiple versions of the image pretty well, and for now I'm just exploring how far it can be pushed before it breaks ... just so I know  

[bigben]

Render time about 26 hours. Not too bad for the render size.

Detail 1, AA 4, no GI
can't remember specs of machine but it's about 3 years old so not the fastest.

[bigben]

Here's a tile from a 3000 pixel wide render. (Damn, you can see the rivers I drew manually) I did a hack of a beach for the lakes and rivers that I won't repeat at this resolution. The water layer was given a 1 pixel stroke around the edge, but this is too wide for the rivers.

At this size you can start to see tree-lined roads and when I fix the grass mask up you'll begin to make out paddocks, although it will require an even larger render size to make them out properly.  You can see even more riverbeds now... I find it quite staggering that it's worked so well for such a quick mask creation.

I think I'll tile the render (so I don't have to breakup the larger terrain) and go a bit bigger again for the final.  One last tile coming, twice the resolution again. I'll use that to see how much work I need to do when I create a new image mask.

[sjefen]

I like your work bigben. This is just stunning.

[bigben]

Thanks.

Here's the tile from a 6000 pixel render from the same camera position (no post work). The mask:render pixel ratio is about 2:1 which is about as big as I'd want to take it without increasing the resoltuion of the mask or using additional blending to disguise the mask borders. There is a fractal child layer providing the colour for each surface but each masked region is quite well-defined as can be seen in close up renders in my masking tests.

I may try a cloud free composite of the final at this resolution. Render times for the tiles are about an hour each, with this res requiring 36 tiles (they will probably take longer near the edge of the globe). I dare say clouds will increase times too much to consider this size (initially).

[SeerBlue]

Have you thought about running your dems thru a gis analysis software such as SAGA Gis (free) to extract masks for the rivers and such, with some experimenting you can get some pretty defined masks. Though I don't know if you can get down to 1 pixel streams, perhaps by running it thru an  colour adjust shader.SeerBlue

[bigben]

It's part of my job to find ways to speed up the tedious 

The terrain is only 90m SRTM3 so the resolution isn't really that useful for really clever stuff, especially in flatter parts of the terrain. There are also several issues with landsat images (JPEG compression artefacts, inconsistent sample times, terrain mis-alignment etc...) which limit how clever you can be choosing rivers "automtaically". The biggest killer is the false colour which will put blue/black pixels in other areas making a colour-based mask unusable.

With higher resolution terrains and full res landsat images I have been able to use a combination of slope map as a layer mask and magic wand selection in Photoshop to quickly select 95% of most rivers. The extra 5% require manually drawing the river to tidy things up, but the results are well worth the effort. With the higher res images you also get more variation in the river width, which can give a river it's own unique charactersitics.

In this case it's going to be manually tracing rivers with a 1 pixel pencil, but as it's only a low res hack I won't have to worry about getting to tricky with the shape of the river.  I may do a full res mask of a smaller area for demo purposes.

In my Grampians project the hardest mask to generate was (and will be) a mask for bare rock. In that specific situation the bare rock is the same colour as lush grass and burnt out forest in the landsat image and drawing a mask that looks good up (relatively) close is extremely difficult.  Editing a 3D mask in 2D is really hard 

That said, it's still worth a look as I'm sure I could find a use for it, so thanks for the suggestion. I did download SAGA once but never got around to using it.  I should also have a look at what else I can do with GlobalMapper now that I've upgraded to the new version.

[bigben]

OK, the last one for this file... double the resolution again (that's the first render at 12,000 pixels wide!). For me this is pushing the limit a little too far for this setup, but I had to find out where it was

Not bad if you don't look too close... but you can start to see just how basic the colouring is. There's 8 main colours on the surfaces with very little variation and bugger all blending. I already have an improved mask for lakes (http://forums.planetside.co.uk/index.php?topic=2519.msg25355#msg25355) which alone would make a substantial improvement, but I'll wait until I have finished the new set of masks before posting more. There is some potential for adding some extra detail inthe shallow water of the sea but this will won't be possible to do around the entire coastline due to clouds and reflections in the landsat images. The middle east coast has a lot of potential though.

I'll give the grass surfaces the same treatment as the lakes which should look awesome and then look at fixing the other surfaces.

[rcallicotte]

To quote Gomer Pyle, "G-o-o-o-lly!"

[zionner]

Wow!

Nice Man! I love All these!

Can I ask...did you use a DEM for this?

If you did..What one? cause i've been wanting to do some NZ work myself.