3DTrains - Add-ons for Microsoft Train Simulator®

Home

Products

Downloads

Help Guides

Support

Help Guides

Help guides

Color reference charts

RailFonts: tips and tricks

Painting guide

Railroad standards

Object Appearance

This will hopefully give some insight into how to use MIP levels to your advantage to achieve the best possible appearance for objects in Microsoft Train Simulator 1.0. While this is far from a complete explanation, it will hopefully guide you in the right direction for learning how to determine what settings to use and how to manipulate shape files and textures.

Scott Miller (Click here for Scott's original guide)

Some background info...

MIP is short for the Latin Multum In Parvo, which means 'many things in a small place'. It refers to a sampled-down version of an image. Kuju .ace files have built-in MIP maps (which I may refer to as mips, because it's easier). What this means is that one 'image file' actually contains more than 'image'. A 256x256 pixel .ace file not only has the 256x256, but also every power of two down to 1x1 - 256, 128, 64, 32, 16, 8, 4, 2, and 1. That's nine total. (see below).

It gets a little more complicated involving DXT1 and .ace files, but I won't go into that here. So why do we care? The Kuju-supplied makeace.exe and makeacewin.exe do not generate mips properly. Mips are normally generated by taking a cluster of four pixels (in a 2x2 arrangement) and averaging them into one pixel (called a box average). This reduces the image by a power of 2. The faulty algorithm used by makeace/win merely chooses a predefined pixel (bottom left, but not certain) out of the cluster and uses that for the pixel in the next mip. While not obvious in the first mip, you can imagine how the error compounds.

images ( 2
	image ( acefile1.ace )
	image ( acefile2.ace )
)
textures ( 2
	texture ( 0 0 0 ff000000 )
	texture ( 1 0 -1 ff000000 )
)

Let's go over a couple of settings first. In the .s (shape) file, there is a group called 'images' that lists all the image files used on the shape. The images are indexed starting at 0 and counting down. Immediately following it is a group called 'textures' which defines texture settings for the images listed. Looking at the example, acefile1.ace and acefile2.ace are images 0 and 1 respectively. You can see that the textures reference this by the first value in those entries.

The other value of importance in the textures section is the third one, which represents the mipmap LOD bias. This basically tells the sim where mips should be activated. A value of 0 means 'no change'. This is what is shown for image 0. Values less than 0 (usually integers, but not required) tell the sim to push the current mip back that much farther. For example, a value of -1 (image 1) tells the sim to use mip 0 (the main image) out to where it normally would, and continue using it out to where mip 1 would normally end, and then switch to mip 1 where it would normally switch to mip 2, switch to mip 2 where it'd switch to mip 3, and so on. Values greater than 0 do the opposite - they switch in smaller mips for larger ones. In general, this is not a concern, because it is easier to reduce the texture size by 50% and save on resources than to have it present, and not use the largest mip. If you're still awake, let's get on with the show.

Note: all the images below are thumbnails. The effects I describe don't show up in the thumbnails very well. Click on them to get the full effect.

Object: default textures, 0 mip bias

So you've produced an object and it's ready for use in MSTS, but it just doesn't look right. What can you do about it?

Shown here, the object has default textures (as-produced by makeace). The mip bias for all images is set at 0. Note the strange effects occurring on the edges of the building. This is the result of makeace's faulty algorithm. Since the texture applied is a repetitive one (brick), strange scaling effects can appear when the pattern repeats in pixel lengths that aren't divisible by two. To put it simply, what happens is every so often, the mortar joint falls on one of the rows of pixels slated for removal, and is disappears. This effectively joins two bricks together. It may not happen on the first mip, but it usually does happen eventually.

Object: default textures, -3 mip bias

I know how to fix it!

Before correcting mip levels was an option, the common solution was (and unfortunately still is, which is part of the reason for this tutorial) to set the mip bias to -3 to basically force the sim to not use the mips until the object would be so far away that you couldn't tell they were messed up. You can see the difference where the strange looking areas are gone. Unfortunately, we're left with another nasty effect. Moiré. These concentric patterns occur because the sim is trying to draw a texture onto a surface that is too small. It has to throw out some pixels, and which ones it throws out depends on the size of the surface. Moiré looks bad enough in the picture - it is much worse while you are moving relative to the object, as the Moiré pattern changes as your relative position changes.

Object: remipped textures, -3 mip bias

So what was this talk about correcting mip levels good for? My object still looks funky!

Just remipping the textures is not enough. As you can see, using remipped textures with the mip bias still set to -3 results in a remarkably similar appearance to the default textures. This means that suppressing the mip levels worked.

This is not what we want. Instead of suppressing them, we want to use them to our advantage. Since the mips are smaller than the full image, they will fit onto the surface better, reducing Moiré.

Object: remipped textures, -1 mip bias

That looks better...

Here, the mip bias for the important textures is dropped down to -1. You can see how the Moiré is reduced, but some still appears on the close side of the wall more perpendicular.

So why the division on a flat surface? The mips kick in based on many factors. One of them being distance from the view source, and another being the surface's angle relative to the view source. It just happens that this criteria was met along that surface, and a division occurred. There are ways to reduce this effect by choosing different texture filters, which I'll discuss shortly.

Object: remipped textures, 0 mip bias

Almost there...

Now we've dropped the mip bias back all the way to 0. It's important to try different options, because factors like size of the texture relative to the object, size of the object, the distance that the object will be from the typical viewing position and other factors all play a part in what will look best. It is important to change the settings, and then move around the object in a manner consistent with its intended purpose to get a feel for how it looks. And from what I've noticed, looking at it in the RE doesn't count. I chose to drop down to 0 because the object is fairly large, and is intended to be viewed from a bit of a distance. This cut down on Moiré at the expense of introducing a little fuzziness, which happens when too small of an image is applied to a surface, which is typically the result of too small of a mip being selected.

Object: remipped textures, 0 mip bias, named_filter_mode changed to LinearMipNearest

default texture, mip bias 0, LinearMipNearest

Good enough. Time for a refreshing beverage of your choosing.

Here's the final result. I also included near and far shots. I reiterate: it is most important to look at the object while some sort of relative motion is going on. This is the best way to choose. In the near shot, you can still see some Moiré, but this can be considered unimportant because you're never supposed to actually be sitting in the parking lot.

Now I'll talk about the texture filter modes. These are designated 'named_filter_modes' in the shape file and are found near the top. Until now, this object had been using MipLinear. Here are the available modes and their descriptions:

Point	No filtering or mip-mapping.
Linear	Bi-linear filtering with no mip-mapping.
MipPoint	No filtering with mip-mapping.
MipLinear	Bi-linear filtering with mip-mapping.
LinearMipNearest	No filtering with filtered mip-mapping.
LinearMipLinear	Bi-linear filtering with filtered mip-mapping.

I don't entirely understand bilinear filtering, nor will I pretend to. In practice, I usually only test the last four and pick whichever looks best.

In this case, I liked how the LinearMipNearest looked. I do know that the difference in 'filtered mip-mapping' versus plain ole' 'mip-mapping' is that it blends the edges of the mips, which would cut down on the noticeable division that we saw in the fourth shot.

So that's part of the story. Now, if you've got a good enough video card, go figure out how to turn on anisotropic filtering and anti-aliasing to improve the visual quality even more.

email: smmille1( AT )yahoo.com