Ron Parker wrote:
> 
> On Mon, 10 Jan 2000 22:25:48 -0600, Mark Gordon wrote:
> >There are two ways I do this: rendering a bunch of stuff with low
> >resolution just to see whether it works (I've known the render to fail,
> >and I've known the render to hang).  Running allscene.sh on my machine
> >typically takes around 25 minutes.  Other methods involve rendering at
> >larger resolution (over a period of days) and visually eyeballing for
> >unusual artifacts.  I've seen all the scenes in the standard
> >distribution enough to have some idea if something goes completely
> >screwy, and I also have "canonical" high-resolution versions saved on my
> >hard drive.  The MegaPov example scenes are not so familiar.
> 
> The reason I asked is because it's something I've thought about myself.
> I think it should be relatively easy to work out an automated solution
> for the cases where there aren't any failures:
> 
> Use pnminvert to invert the canonical copy of the image, then use ppmmix
> to mix 50% of it with 50% of the test image.  Finally, run ppmhist on
> the result and examine its output to see how many pixels are significantly
> different from solid 50% gray.  If it's over a certain threshold, you can
> guarantee there's been a significant change in how the image was rendered.
> 
> Adjusting the threshold and the definition of "significantly different from
> solid 50% gray" allows you to detect smaller and smaller differences in
> rendering, up to and including perfect reproduction of the original image.
> So you can sort the images into categories: "really different - probably a bug",
> "A little different", and "perfect".

I had been picturing something similar (though I was going to use the
GIMP), but I haven't gotten around to it yet.

-Mark Gordon

Post a reply to this message