Here is a simple scenario that illustrates why taking any model or batch of models from CEP (or anywhere else) and just running them through CleanModels is going to risk being a boondoggle. This is not, in any way, meant to undermine the usefulness of CleanModels. It can be a very useful tool, but it's not necessarily useful in all situations and, frankly, the results from CleanModels can be very misleading in evaluating the quality of a model.
This misleading evaluation thing? It's not CleanModels or OldMansBeard's fault. This is a perceptual flaw on the part of many of its users. It's a
very easy mistake to make.
1. Make simple teapot placeable. Here's how I did that, and the rest on this line is only important if you want to reproduce my work: Make an aurorabase. Make a teapot in GMax (Radius:250, Segments 4). Move to X, Y 0.0, Z 250.0. Drag and drop a texture on it. Convert the teapot to editable mesh. Add an Aurora Trimesh modifer with whatever settings you like. I use Render and Override Material Values only. Attach it to the model base and export. I export geometry only, in this case.
2. Run it through CleanModels. Outside of the default settings, I'm doing a binary snap because this model is headed for the compiler when it's done.
Result: plc_a01.mdl loaded.
binary snap applied to node positions and mesh vertices
vertices welded in [teapot01]
null faces deleted from [teapot01]
unused tverts deleted from [teapot01]
tverts welded in [teapot01]
renormalized tverts in [teapot01]
Fixes made = 11073
plc_a01.mdl written.
Total Fixes =
11073Okay, 11,000 fixes. Are there
really 11,000 errors that need fixing in that simple placeable? No, of course not. Are there 11,000 operations that CleanModels has carried out against that model which were at variance with the settings it was run with? Oh yes, I have little doubt of that.
However, 11,000 CM3 fixes ≠11,000 actual errors of signifigance.Are
some of those errors signifigant? I'm sure, in some way, they are or could be argued to be. Are those errors noticable in-game and/or do they have a negative impact on the Aurora engine? That is more up for debate, IMO. But that's far from the point I'm trying to make, which can only be evinced by compiling these models and having CleanModels analyze them again.
Before we do that, though, let's take the outputed model from CM3 and run it through CM3
again, with the same settings. Looks like there are still some errors left, according to CM3:
plc_a01.mdl loaded.
binary snap applied to node positions and mesh vertices
renormalized tverts in [teapot01]
Fixes made = 444
plc_a01.mdl written.
Total Fixes =
444Again are there really 444 errors
of signifigance in this model? No, there aren't. You can keep taking the outputted model and feeding it back into CM3 with the same settings and you're going to get a notification that there were 444 fixes performed
each time.
Even all this so far, as I said, is not my main point. Let's take that model, that CleanModels produced, and compile it with various compilers and then have CleanModels analyze
those binarized models:
Internal BioWare compiler results: Total Fixes =
9271DLA Model Compiler: Total Fixes =
9271Acaos mdlcomp from Explorer Reborn 1.63: Total Fixes =
9271After compilation, all three are reported by CleanModels to have gone up
20x the number of errors that needed to be fixed. Interestingly, all by the same amount.
What's actually going on here? What does all this mean?
It means: If you load a model into CleanModels on the hope of looking for errors, you will likely not be disappointed. You will likely see hundreds if not thousands of errors, just like you expected. Using those numbers to support any kind of assessment of the quality of the model is often specious. Only a relatively skilled modeler can actually make that assessment, especially with complex geometry. The teapot placeable I made compiled just fine with the Bioware model compiler and worked fine in-game.With the NWN2 models, and with basically anything else in the existing CEP package, I think the main questions should be:
* Is this model
reported to have a specific error of some sort in the first place?
If so:
* Does the model compile well?
* Does it look good in game?
* Is there anything about it that taxes the engine, unnecessarily, and needs to be optimized?
CM3 is a powerful and useful tool, but feeding models into CM3 can lead to unintentional trips down the rabbit hole if one isn't careful. Worse, its output can be misinterpreted to call into question the quality of models which are, at least in this case, more or less perfectly fine.
Modifié par OldTimeRadio, 27 janvier 2014 - 11:27 .
