*Exhibition.
Ahh... every time I come through Tech-Related I get a bit sad about my inability to rip things like some of the rest of you... I have so many games I want investigated, but no time, or skill, necessary.
Koral and MrAdults specifically, you two are my model decoding heroes. I've never seen anyone else get so much done so fast. Is there a secret you have? Like being aliens?
I used to wonder how people did it, too, back in my early programming days.
There's no real trick, it's just a matter of having a wide base of knowledge to narrow down the possibilities. If you know what you're looking for before you start looking, you'll find it faster. On the other hand, you can't be too focussed on a single possibility, or you'll waste a lot of time looking for the wrong thing.
Some practical examples on what to look for in terms of vertex or triangle data (assuming your data set is not compressed/encrypted to start out with, which is another different set of issues):
Usually you can find triangle lists by searching a file in 1-byte-offsets for shorts or ints backed up against each other, all within a specific sanity range (where the maximum value would be the highest vertex index). If that fails (triangle indices may not be back-to-back) you can can try searching for specific geometry bits in different pattern orders, like groups of 3/4 indices. The list goes on into all kinds of different and weird/more obscure indexing modes.
Vertices are usually trickier. If UV's or positions are in floating point, it's pretty easy. Just search for valid floating point value ranges in a specific stride, or back-to-back in the case that each vertex component is in its own array. Bytes, shorts, and ints, on the other hand, can blend in with all kinds of data when utilizing their full ranges. In that case, you can try some patterns based on known index values, or if you don't have known index values, you can check value ranges and failures to comply with those value ranges outside of a given stride. If all of that fails, you usually have to start manually looking through the hex and interpreting values, and just trying to find patterns and/or consistently recognizable values in the interpreted data. The biggest thing to know, regarding the whole picture, is that the more pieces of a file you document, the easier it becomes to narrow down the rest. Since if you already know what some piece of data is, it can't be the other piece of data you're looking for.
Then outside of scanning for raw values, you have cases where there is some obvious header data with offsets to specific pieces of data (like the Crisis Core models), or cases where the file is broken up into chunks with a common chunk header, like the Dissidia/GMO models. Those things are pretty easily visible by interpreting the data as raw shorts/ints/etc. and just taking a look at the kind of data that's at the offsets (finding offsets is, I guess, mostly a matter of intuition and trial-and-error - if it's in a storage type that can hold up to the filesize, and the value is less than the number of bytes in the file, then maybe it's an offset
). If you're lucky enough to have that kind of data at your disposal, it's much easier to interpret each specific chunk/offset of the file for a specific type of data, rather than having to bruteforce through the whole thing. On the other hand, there are also much more complicated methods of breaking files up into sections, and games will often have complete sub-filesystems within files for given types of files, or bits of files that are compressed, or all kinds of other things. Dealing with all of the scenarios just requires a comprehensive base of knowledge, so it's one of those things that you just get better at doing with the more you learn about storage techniques and other areas of development (for mesh data specifically, knowledge of rendering, more-so of rendering on the model's target platform(s), is most useful).
I guess I'm getting a bit off-topic there. I just wanted to take the opportunity to de-mystify the process a little bit for anyone interested.