How can they solve the massive texture problems?

[AwesomeOcelot]

IceWind Dale 2 is more modern than Fallout, Fallout 2 or Diablo II.

So point doesn't stand.

 

What point was that? Games released later take advantage of newer technology. It's only a fair comparison if they're released around the same time. If Blizzard had made Diablo 2 around the time of Icewind Dale 2 it would have looked much better.

 

I kinda expect some re-use of smaller areas (think housing in Baldur's Gate 1), however areas should have their unique feel, as do cities, as do dungeons. You just don't get that with tiles.

So, no thanks. Not sure why anyone would want that when we can have fully handcrafted areas instead. And then you rather want tiling?

 

Games with tiles like Diablo 2 do that, each chapter has different tile sets and a completely different environment, plus many different smaller areas with even more different looks, and also unique rooms and objects. Each area can have unique tile sets, and that still uses less space and requires less work than creating every area.

 

a) We can't have fully hand crafted areas at the quality of that screen shot with the amount of time and size of the team.

 

b) Because tiles are way more efficient, more time can be spent on all the other aspects. It's also flexible how much repetition there is.

[Teahive]

Infinity engine did in fact support tiles. Not everything in those games is unique.

 

Let's start with a typical area - something like this (which was posted in a different thread by D3xter:

 

This is a 3840x2880 full area from IWD. Now, PE will have higher resolution areas .... so lets say that a comparable area in PE would be 10860x8140, which means around 8 times more pixels. Fair enough? Now, this would mean 88,400,400 pixels. with 24-bit colours+alpha channel , the uncompressed size would be 336MB. It should be possible to compress that fairly easily down to 100MB or so ... and with lossy compression down to 45MB or so.

No need for guessing, the IWD image is a 4.13MiB JPEG image, which means it's a little more than 3 bits per pixel, or 13% of an uncompressed 24bpp image. And it's still very high quality. So yes, 45MiB of on-disk storage for 8 times the pixels would be plenty.

 

With optimised compression you could probably get below 2bpp.

[Frisk]

 

b) Because tiles are way more efficient, more time can be spent on all the other aspects. It's also flexible how much repetition there is.

 

I think you are completely missing the point. The game is not going to use tiles, but that does not mean that they have to do every area from scratch - there will be reusable resources. The can re-use things like trees and such as animated "placeables". Alse, when the build the model for each area, they can re-use components - walls, buildings, statues, etc that they have used elsewhere - place them in the model and render them from different angles than they did before.

 

There are numerous other (and better) ways to re-use resources than to use tiles.

[AwesomeOcelot]

b) Because tiles are way more efficient, more time can be spent on all the other aspects. It's also flexible how much repetition there is.

 

I think you are completely missing the point. The game is not going to use tiles, but that does not mean that they have to do every area from scratch - there will be reusable resources. The can re-use things like trees and such as animated "placeables". Alse, when the build the model for each area, they can re-use components - walls, buildings, statues, etc that they have used elsewhere - place them in the model and render them from different angles than they did before.

 

There are numerous other (and better) ways to re-use resources than to use tiles.

 

No, you're completely missing the point of the advantages of tile based systems. You'd have animated placeables in a tile based system. You also reuse assets from one tile to create another. It still takes more work to create these backgrounds than a tile based system. If they had to rotate a 3D model, they'd have to render the scene, paint over it, create pathing and lighting maps. Unlike a tile based system, with these systems you can't just drop in a building, a bridge, that's already rendered, already painted, already has pathing and lighting maps.

[mstark]

It still takes more work to create these backgrounds than a tile based system. If they had to rotate a 3D model, they'd have to render the scene, paint over it, create pathing and lighting maps. Unlike a tile based system, with these systems you can't just drop in a building, a bridge, that's already rendered, already painted, already has pathing and lighting maps.

I would much prefer to have a limited set of completely unique maps, as compared to the very same, finished structure/statue/patch of grass pasted over and over in different areas of the game. This is exactly why I hated NWN, and completely ruins immersion.

 

If they want to create a village with similar houses as to another area they can copy the 3D assets, change some details up, and render it again. That would at least save the time of creating entirely new models, since rendering can be done without using up man hours.

 

IE games used quite a bit of "tiling", especially in corridor based areas. Didn't like it. I wouldn't mind if they made a set of reusable models for torches, chains, and other detail items, and then reuse them all over the game. It kind of irks me to see the same set of torches & torture equipment in every single dungeon, though. If it can all be made unique, I'd take quality over quantity any day with a game like this.

[Pipyui]

Alright, so I read a bit more into tiling, and came to my own conclusion that tiling is a dead end for PE if we want anything remotely close to unique landscapes.

 

So that brings us (or me anyway) back to where we started. Unique landscapes require, according to the devs, a whole ton of memory. Will compression be enough, or will other tricks, less lossy than tiling, be necessary? What kind of compression should be used, anyway?

 

Edit: Sorry for the self-righteous attitude here, just trying to get a little more inspiration in this thread - the tiling debate is going nowhere fast.

Edited October 26, 2012 by Pipyui

[AwesomeOcelot]

Infinity engine did in fact support tiles. Not everything in those games is unique.

IE games used quite a bit of "tiling", especially in corridor based areas. Didn't like it. I wouldn't mind if they made a set of reusable models for torches, chains, and other detail items, and then reuse them all over the game. It kind of irks me to see the same set of torches & torture equipment in every single dungeon, though. If it can all be made unique, I'd take quality over quantity any day with a game like this.

 

It wasn't a tiling system in the same way Fallout or Diablo are tiling systems. In either system, there's support for as many different detail items as Obsidian want to make. They've already committed to a large world, so they have to work to what ever quality they can to get there. Time and man hours are a concern, regardless of how unique or how much quality you want.

[Sensuki]

The game won't use tiles, but they are building levels in 3D and then rendering them as a 2D image so before they render it they may re-use 3D assets (like Baldur's Gate 1 did with the trees and houses), which is completely fine with me.

 

They don't need tiles to re-use content.

Edited October 26, 2012 by Sensuki

[Infinitron]

but...but...lighting pathing painting rendering!

Edited October 26, 2012 by Infinitron

[Hassat Hunter]

Unique landscapes require, according to the devs, a whole ton of memory.

To create on their side, yes. Not for the enduser to display them and properly run them (after all, it'll just be a bitmap).

 

But with the current memory prices, windows allowing a lot more than in 2000 and all that jazz, it's definitely less an issue than it was in IE's area.

They already stated themselves they no longer need days for rendering, rather hours... so...

[Nightshape]

Infinity engine did in fact support tiles. Not everything in those games is unique.

 

Let's start with a typical area - something like this (which was posted in a different thread by D3xter:

 

This is a 3840x2880 full area from IWD. Now, PE will have higher resolution areas .... so lets say that a comparable area in PE would be 10860x8140, which means around 8 times more pixels. Fair enough? Now, this would mean 88,400,400 pixels. with 24-bit colours+alpha channel , the uncompressed size would be 336MB. It should be possible to compress that fairly easily down to 100MB or so ... and with lossy compression down to 45MB or so.

No need for guessing, the IWD image is a 4.13MiB JPEG image, which means it's a little more than 3 bits per pixel, or 13% of an uncompressed 24bpp image. And it's still very high quality. So yes, 45MiB of on-disk storage for 8 times the pixels would be plenty.

 

With optimised compression you could probably get below 2bpp.

 

Lossy compression is not the way to go at all...

[rjshae]

Then use non-periodic tiling!

 

Another approach is fractals. I know they have fractal-based systems for generating terrain. Somebody must have done comparable work with fractal texture generation. That may be useful for generating consistent, non-repeating terrain patterns, such as grasslands or desert sand. Basically trading disk memory for CPU load.

[AwesomeOcelot]

Then use non-periodic tiling!

 

Another approach is fractals. I know they have fractal-based systems for generating terrain. Somebody must have done comparable work with fractal texture generation. That may be useful for generating consistent, non-repeating terrain patterns, such as grasslands or desert sand. Basically trading disk memory for CPU load.

 

Fractal-based, non-repeating? I thought fractals were about a pattern repeating itself so it looks the same close up or far away. It's the way nature repeats itself, so it makes sense to have fractal patterns in tile systems of nature.

Edited October 26, 2012 by AwesomeOcelot

[john_boy]

Im still a bit concerned on how large the download will be, that is all

 

If it could be split into chapters, a download for each - that may help, if it's viable at all.

[Archmage Silver]

I agree that we want a PC only game, Im still a bit concerned on how large the download will be, that is all

If you were to get your copy at Goold Old Games, you could use Free Download Manager or equivalent to download some of the game installer, pause, and then continue later. If people are facing a download limit per month, this could be worth it if done at the end of the month, which is when that download limit would reset to zero for the next month, soon after the partial download.

[PIPBoy3000]

This has been an interesting discussion to follow. I suspect that Obsidian plans on creating very large textures as part of the art process, and then will size it down according to various technology limitations (bandwidth, DVDs, compression, etc.). There will also be an upper "art" limit where the artists take the high-resolution render and apply their touch-up work - scaling it up from this art limit won't produce any more detail. I'm guessing they have some resolutions in mind, but might work at twice that resolution to give themselves some breathing room should things change.

 

I suspect that even if the DVD and standard download package contains somewhat low-res backgrounds, they'll give people the opportunity to download higher resolution backgrounds if they'd like.

[rjshae]

Then use non-periodic tiling!

 

Another approach is fractals. I know they have fractal-based systems for generating terrain. Somebody must have done comparable work with fractal texture generation. That may be useful for generating consistent, non-repeating terrain patterns, such as grasslands or desert sand. Basically trading disk memory for CPU load.

 

Fractal-based, non-repeating? I thought fractals were about a pattern repeating itself so it looks the same close up or far away. It's the way nature repeats itself, so it makes sense to have fractal patterns in tile systems of nature.

Yes they are self-similar, but that doesn't mean they must produce identical, repeating patterns like tiles. For example, you can use fractal algorithms to generate any number of trees, no two of which are the same. But it's probably too much work to be worth implement anyway, unless there's a graphics library handy.

Edited October 26, 2012 by rjshae

[Kuroneko]

There are algorithms to encode arbitrary images as fractals/iterated function systems. They recursively look for similarities between different portions of an image and encode a mathematical function such that iteratively applying it reproduces the original image.

 

The result is extremely space-efficient, or at the same compression factor much better quality than jpeg/etc. Because it's encoded in terms of geometrical operations, it can be decoded at any resolution, with the resulting image typically looking a lot better than ordinary interpolation methods.

 

Too bad that last I heard its licensing is pretty draconian.

[Rokku]

The whole reason this has been raised as an issue is because of how distinct and unique the released image was. Tiling or fractal compression just doesn't make sense for hand crafted non repeating imagery. Sure maybe a rock here or a tree there but such props are only a minor component of most scenes and might result in single digit compression improvements.

[Kuroneko]

You're mistaken about what fractal compression is. It does not at all require literally repeating imagery, nor is that its normal use. For example, it was used in a version Microsoft's Encarta encyclopedia for its images in order to fit it on a single CD. Typically it's used to pack images into very tiny filesizes (and hence quite lossy, though typically better than comparably-sized alternatives), but it works for any level of detail, so it doesn't need to be that extreme. There's even a Photoshop plugin that uses it to resize images, as an alternative to bilinear/bicubic/etc. interpolation.

 

Fractal compression is very slow to encode, but very fast to decode, and it works for practically any kind of image. I suspect that the real downside is that the good automated algorithms would likely make Obsidian pay through the nose in licensing and be complicated to implement, so just going with normal methods and swallowing the file sizes may work better in the end.

[Rokku]

Ohh I didn't know that my mistake.

 

Yeah in that case its sounds pretty good I just wander what the compression ratio would be for PE style images.

[Teahive]

Lossy compression is not the way to go at all...

Please explain. Given the limitations of the human visual system there is no reason not to go lossy. You simply need to make sure not to go overboard, and no one will be able to tell the difference.

[rjshae]

Lossy compression is not the way to go at all...

Please explain. Given the limitations of the human visual system there is no reason not to go lossy. You simply need to make sure not to go overboard, and no one will be able to tell the difference.

They could perhaps do a hybrid approach: lossy compression for the background, accompanied by non- or low-lossy overlays for regions that display obvious compression artifacts.

[Adhin]

Well I went ahead and actually made a bmp the same size as some... guy here mentioned, 10860x8140. BMP, no special compression, 24bit, 252 megs. I can see 32 bit being his actual projected 350-ish. Either way that's pretty high up there with out any major compression going on and with todays compression you can get BG2 or BG1 onto a 1-2 CD's, instead of the 4-5 they had to use back then. And that was a good 5 or so years ago, its gotten excessively better since then.

 

As for effects like shadows, special lighting outside of the pre-rendered stuff to the image its self... none of that's texture of major print on the games size, its all overlay renders... pixel shader stuff, it happens on render. Same with water in games, and I doubt they'll do a high frame count repeat water texture for there water. It'll probably end up being a 2nd layer, painted on (like drawing line around the 'water' areas) that do the pixel shader driven water stuff like... every games done since Morrowind and a few before.

 

What im getting down to is I'd imagine the download print would end up being around 15gigs unless they go well beyond a 50 map mark. To give you an idea, going off uncompressed images of 250 per, 50 would be 12.2 gigs. Any real compression would drag that waaaay down to ehh, well a JPG of the same (they wont use that but its an example) is about... 84MB per map instead of 250 aand presto, 4.1 gigs. If this games 15+ gigs after compression its going to be deliciously giant, seriously 'giant'.

[Rokku]

Lossy compression is not the way to go at all...

Please explain. Given the limitations of the human visual system there is no reason not to go lossy. You simply need to make sure not to go overboard, and no one will be able to tell the difference.

 

Even if you were unable to see a major difference with your eyes using lossy compression it becomes an issue when you start lighting special effects. This might introduce artefacts that you can see...