[Plugin] OBJexporter v3.0 20130131

Sound

Thank you for the answer. Yes, it looks like those internal faces were already in SketchUp. A new test shows 12 facets, and that's fine.

Still, the other problem persists. Even when the exported cube results in 12 facets, the file contains 24 vertices (you can check this by opening the OBJ file with Meshlab or any other importer, or even by counting the "v" lines in the OBJ file). This looks very wrong, as the original geometry only has 8 vertices. Can you help me about this issue too? Thanks!

TIG

So no apology for putting me to unpaid trouble when it was entirely your error?
My exporter does work differently from the pro-OBJexporter, which minimizes the number of vertices by recycling them when possible.
However, my version treats each context, material and face as a separate set, and doesn't reused vertices. The coding is such that it would be very awkward to effect...
However, it has no noticeable effect on the resulting form and a ood 3rd party app should be able to cope with welding vertices etc anyway...
Get Pro if you really do NEED this functionality, and are not just being 'anally retentive'.
After all my tool is 'free' [donationware] and you cannot readily expect, let alone demand, changes to its functions...

Sound

Well, I'm just reporting a bug. A cube, having 8 vertices, is exported with 24 vertices
This is a wrong representation of the object and has a very noticeable impact on all uses of the OBJ file when topologic integrity is required (3D printing is one example).

Of course I expect nothing from you, being it freeware. Whether you want to fix this bug or leave it there is entirely up to you.

TIG

@Sound

The vertex 'duplication' isn't actually a 'bug', since that would suggest that it isn't doing what it is intended to do: and it currently does exactly what it is coded to do.
However, I do take your [inelegantly put] point that vertex entries could be 'reused' if they are 'shared' by more that one facet...

So I now have an idea on to make an adjusted version of the code that will now does this: and it can also have a few other tweaks; to reduce the file size further, to optimize the processing time etc.

I'll post it soonest...

PS: Although I do 'reserve all rights' to my code... please feel free to make your own improvements to any of my tools, and either PM me or post them here, of course for free unfettered use by all members... The only thing I ask is not to post whole scripts in the same thread as it'll cause confusion...

TIG

Here's v2.8 http://sketchucation.com/forums/viewtopic.php?p=294844#p294844
The D.P. accuracy is not now applied to 'whole numbers', and any -0 values becomes 0.
Any shared-vertex entries are now 'reused'.
The OBJ file size is minimized and the processing time is optimized.

Sound

TIG, thank you very much for working on that and for clarifying your terms about modifying the script.
I will test the new version as soon as possible and report success or failure.
Have a good day.

Sound

I can confirm this works for any model I tried.
Thank you! Exported OBJ is now topologically correct and thus suitable for a lot more uses.

TIG

Good.

TIG

Here's v2.9 http://sketchucation.com/forums/viewtopic.php?p=294844#p294844
Merging of shared vertices is now optional, as it slows down processing for larger models considerably; and it is only required by some more picky mesh-apps, but renderers etc don't seem to mind non-merged vertex data at all.

TIG

Here's v3.0 http://sketchucation.com/forums/viewtopic.php?p=294844#p294844
It has a much improved shared-vertex merging algorithm that is now slightly faster than a non-merging version, so the option 'not to merge' has been removed because it is now of no benefit.

Where possible, shared-vertex data is always reused, ensuring correctly formed geometry.
This can mean that a SKP export containing 100,000 faces and therefore potentially 300,000 vertices can be 'compacted' to use perhaps 50,000 shared-vertices [but of course this depends on each SKP's geometry, grouping etc]: this 'compaction' also reduces processing time [which will now be perhaps ~60 seconds for this sized example], and when combined with the latest optimized 'numerical-entries' gives much smaller OBJ file sizes, and faster processing in 3rd-party apps too.

andybot

Great work, as usual, TIG!! This plugin is very essential for me. Many thanks for your development efforts!

cotty

Thank you for this update!

samurai1789

What simple instructions???? What ever i do i cannot find it HELP!!!!

TIG

@samurai1789 said:

What simple instructions???? What ever i do i cannot find it HELP!!!!
They are on the download page ! http://sketchucation.com/forums/viewtopic.php?p=294844#p294844
There are also some simple dialogs as you go after the tool is started...
What problems do you have otherwise ?

thomthom

@tig said:

The D.P. accuracy is not now applied to 'whole numbers', and any -0 values becomes 0.

What made you chose to do this? Are there applications that doesn't handle these values well? I'm curious because I've been looking at the OBJ format this weekend.
(Is there an official specification somewhere? I find multiple sources around...)

pilou

See the external links at the end of the page

thomthom

@unknownuser said:

See the external links at the end of the page

Is that the official specs though? "martinreddy.net" ...

TIG

@thomthom said:

@tig said:

The D.P. accuracy is not now applied to 'whole numbers', and any -0 values becomes 0.

What made you chose to do this? Are there applications that doesn't handle these values well? I'm curious because I've been looking at the OBJ format this weekend.
(Is there an official specification somewhere? I find multiple sources around...)
It just minimizes the OBJ file size, an importer reads 0 as 0.0, so making it 0.000000 is pointless, -0==0 so it might as well be 0 !... so all trailing zeros are now omitted, it speeds up the writing by a few % and I also the importing at the 3rd party end...
There's an OBJ Wavefront Wiki...

thomthom

Are you sure all importers manage to read an int instead of a float which the specs appear to indicate? Just wondering in case the parsing is implemented in a stricter language...

@tig said:

I also the importing at the 3rd party end...

Something missing from that sentence?

@tig said:

There's an OBJ Wavefront Wiki...

This one? http://en.wikipedia.org/wiki/Wavefront_.obj_file

TIG

I have used OBJ files exported like this in many apps [Octane, Blender, MeshLab, MeshMan, AC3D etc] and they all import just fine.
As far as I can tell they expect a 'number', so 0 == 0.0 == 0.000000 etc, 1 == 1.0 == 1.000000...
They even take -0 == 0 when -0 is not really a number.
I also trap for 'NaN' and 'Infinity' which would break them...
Sharing vertex data is also permissible if you correctly reference the vertices.

Try it yourself using a box that is located at, and offset from, the ORIGIN by whole 'meter' distances. The native OBJ exporter using m/triangulation, flipped-YZ etc gives a similar result to my 'free' tool BUT 1.000000 >> 1 in mine... Now import the two OBJ files into an app like Blender and you get the exact same result... If you make the two much more complex you'll find mine loads into the app a tiny % faster. Of course it takes longer to write the file because the native exporter is a swifter exe...