Thanks, that's really helpful! Yes, and I should have mentioned in my original email -- if there is an application that has already added UDIM support that is widely used or that you like its conventions, please point it out and I'll try to use the same notation (if there is any consensus among such apps).
Michel, I'm happy to support the Mudbox tiling as well, can you outline how it is different? > On May 31, 2016, at 2:51 AM, Michel Lerenard <[email protected]> wrote: > > Hi, > > I add support for UDIM in Clarisse over the TextureSystem a few years ago, I > should have a few answers: > > - I used the <UDIM> tag. "%04d" is too generic and may confuse users in my > opinion, since it could also be a way to specify a frame. Having an explicit > tag that stands out is useful. (so we have two tags: <UDIM> for Mari and > <UVTILE> for Mudbox. ) > > - Agree that UDIM is single tile. Tile indice in UDIM is hardcoded in > Clarisse to 1000 + 10 * V + U + 1. Mari seems to be harcoded as well, so > unless you have a in house software package that support other sizes, I don't > think you'll need another line width. We've never had any complaint so far. > Per texture option : useless. > At texture system level: why not, it shouldn't make the code much more > complex. > > - In my implementation, I do not make any assumption on the data: it may or > may not exist for a UV range, resolution may change, so does AOV/channels. > It's not the most efficient way to go, but I'm sure that it will work every > if you mix mipmapped TX files with loads of AOV with simple Jpegs. > > Here are a few thoughts that I'd like to share with you about this feature: > - The main issue I have today is being able to use filtering over several > tiles. I compute the name of the file to evaluate, and offset UVs before > calling the texture() function, which means that OIIO does not know that I'm > evaluating UDIM tiles => I can't get any filtering on the sides of the tiles, > or integrate data over several tiles. Making this work would be awesome. > > - Listing the available tiles for fast access can be easily done. What I do > is parse the folder, get all files matching the filename and create an array > to later be able to evaluate a tile without accessing its filename. Very > efficient especially if you have holes. > > - Do you plan to support Mudbox tiling system as well ? It's not that > different and is able to handle negative ranges, which is very pratical for > symmetry. > > - I'm interested in testing it, i'm curious to see if it's faster than my > implementation, in which case i'd switch. > > > On 05/30/2016 11:41 PM, Larry Gritz wrote: >> I've gotten a few requests lately for direct UDIM support in OIIO (and, >> transitively, in OSL). >> >> The way I figure this will work is that you pass in the UDIM u and v texture >> coordinates (which may extend outside [0,1], with each [i,i+1) block >> indicating a different texture region tile), and for the filename you will >> give a generic name such as "myfile.<UDIM>.tx". For the particular texture >> lookup, the u and v will be assessed and the "<UDIM>" will be replaced with >> the right tile number (let's say "1013" for u=0.25, v=0.12). >> >> Don't worry, I know a way to do this so that there is no actual string >> searching or construction happening per call. So the expense of this will be >> very very low. >> >> But I want to make sure everybody is happy with the way of signalling this, >> so I have a few questions. >> >> * What text should indicate that it's a UDIM texture and which will be >> substituted with the tile number? "<UDIM>"? "%04d" or some other explicit >> format indicator? Something else? >> >> * Does everybody agree that UDIM tiles are 1-D (i.e. a single tile number, >> not a separate u and v tile in the filename), start with 1001, are 4 digits, >> and always have 10 tiles in the u direction? Can this be hard-coded, or does >> it need to be an option to the ShadingSystem (that could be overridden on a >> per-site/per-product basis), or does it need to be something that can be >> specified separately for every texture? >> >> * Because the filename you pass is "virtual" -- that is, "foo.<UDIM>.tif" >> doesn't actually exist, it's only turned into a proper existing filename in >> the midst of a lookup for a particular u,v -- this means that >> TextureSystem::get_texture_info which lacks uv coordinate parameters must >> necessarily fail for most queries, since you don't know which individual >> texture region file should be used. Does that sound reasonable for those >> queries to fail? >> >> * Are there any properties that you feel strongly should or should not be >> constrained to be identical for all of the individual region files for one >> UDIM texture? I assume we want to allow the region tiles to have differing >> resolutions, say. But should/can we assume that they must all share the same >> number of channels? Are there any cases I need to be aware of where >> differences are nonsensical? Any cases where differences are inevitable and >> I need to ensure that such flexibility is allowed? >> >> Let me know if you have opinions. As well, even if you are fine with the >> proposal but are keen to try out the UDIM support as soon as it's available, >> let me know so I can ping you for feedback when it's ready to take for a >> test drive. >> >> -- lg >> >> >> -- >> Larry Gritz >> [email protected] >> >> >> _______________________________________________ >> Oiio-dev mailing list >> [email protected] >> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org >> > > _______________________________________________ > Oiio-dev mailing list > [email protected] > http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org -- Larry Gritz [email protected] _______________________________________________ Oiio-dev mailing list [email protected] http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
