I think I should turn everyone's attention to a thread on the HPZR forums:
http://hpzr.proboards.com/thread/7484/c ... ting-suite
An excerpt from that thread:
wurlitzerfox wrote:I'm currently working on a sort of hacking utility/editing suite for Crash 1. It's still at an early development stage so I can't release too many details, but as of right now I'm only able to view object models, level models, and disassemble executable GOOL binary code (GOOL is the scripting language that Andy Gavin/Naughty Dog developed for the Crash series; a majority of the game's function is the result of executing 'GOOL processes' for every conceivable game object, which are managed by the game engine in much the same way an operating system manages its processes, where a sort of inter-process communication/event based system allows objects to communicate with one another).
From the screenshots, you can see that there is a significant level of support for extracting level and object models for Crash 1 (although currently untextured). There's also some kind of GOOL bytecode disassembler shown in the screenshots.
This is easily the most significant work done so far (at least that I know of) on reverse-engineering the game. Although most of it applies directly to Crash 1, a lot of it is still very much applicable to Crash 2 and 3.
Anyway, that aside, there have been some stuff:
Entries actually have names, interestingly enough. I just assumed they were identified in an arbitrary numerical manner, but sure enough the "entry ID's" were specially-encoded strings. Credits to wurlitzerfox (from that thread) for finding this one. These strings can sometimes be pretty human-recognizable, so I don't think they are necessarily computer-generated. Names like "WillC", "FruiC", "BoxsC", and "ScrbC" show up for GOOL entries, for example, and they contain the code for objects identified by the same name.
SVTX is apparently the vertices for the object models. The polygon/texture-mapping data is held elsewhere, in the TGEO entries (or so I've come to understand, I don't know the specifics to this one). Presumably, each SVTX would be one frame or keyframe of animation for a given model, considering that texturing does not change between different frames of a given animation. Again credits to wurlitzerfox, he's the guy who's figuring out all this stuff.
As of commit
ddc78cf, CrashEdit supports viewing Crash 1 level scenery (minus the texturing) and exporting to OBJ and COLLADA formats, although blender really doesn't like opening the output models (maybe they're too big or something? I think I should scale them way down). Noesis supports them just fine, though. There's no vertex color support in OBJ, which means all you'll get is vertices and colorless, untextured polygons. It's really unfortunate, because the only format I know of that supports this is COLLADA, and that's a super-complicated (and surprisingly strictly trademarked) mess. I don't think I'm even exporting to it properly, either. Anyway, there's no compiled build for this version of CrashEdit, but I can make one if someone requests.
I've got some part of the Crash 2/3 scenery models down (vertices are good, polygons are not yet understood). Nothing reliable enough so far, though. (EDIT: oh yeah, I found the Z coordinates by the way)
Anyway, that's all I can think of right now. That guy's reverse-engineering progress is amazing. All kinds of great stuff is coming from that.