Hi Seth, Well, you're partly right that you wouldn't need to load all of the scene in to memory. But, don't forget that shadows can be cast from one end of the scene to another. These long range effects would be even more pronounced when reflections were present, like they may be with POV Ray. In that case you certainly wouldn't be able to make an a priori assessment of the part of the scene that could contribute to a certain ray from that part of the viewport. Then again, if you feel that you can cut the scene in parts, you can achieve the same effect by turning off parts of the scene, making them black. Cutting on the number of objects would definitely cut on the complexity of the scene and hence on the memory demand. Subsequently you could stitch the parts of the image together by mere addition (since black pixels count as rgb 0 0 0). But I'd go for hash_max... Well, actually I'd probably go for POV Ray... :p
Cheers, Tsjerk > Hi Tsjerk, > Ok, yes, I see your point now. I was unaware of the memory aspects not being > helped by the rendering in parts, so I did not catch the deliberate > redirection you had done! However, Tom's method would be able to avoid the > too large memory issue since if the viewport only 'saw' one little piece at > a time and worked on that independently of the rest of the scene > (essentially unaware if there is one more tile or a hundred more...) But > there are the multiple perspective limitations, etc. > Thanks for clarifying the 'under the hood' part for me and sorry to have > then added my own assumptions to the mix!! (as I worried about in my last > paragraph!) > -Seth > > On Sat, Oct 24, 2009 at 2:41 AM, Tsjerk Wassenaar <tsje...@gmail.com> wrote: >> >> Hi Seth, >> >> Rendering in parts does not avoid taking the whole scene in memory. >> That is a more important obstacle than the size of the image. That is >> why Warren suggested using the hash_max setting, which will alleviate >> the memory dependence of the ray tracing, avoiding crashing. The >> raytracing in parts that POV ray offers is not about saving memory; a >> scene that is too complex to load will still break. It's about CPU >> efficiency: you can distribute the calculation over multiple CPUs. >> This was particularly important for POV Ray versions <= 3.6 (3.7 will >> be able to use multiple CPUs). Also, there's nothing mathematically >> tricky about what POV Ray does for partial ray tracing: it only traces >> the rays that belong to pixels you want to have rendered. E.g. it will >> write a 9000x6000 pixel image (according to the header), but only have >> part of the image filled in. >> >> Maybe you can see that I wasn't so much missing the point, or maybe >> was deliberately trying to miss it, since it builds on misconception >> of what is under the hood. If you want to connect withe the >> capabilities that POV ray offers in that regard, which may have some >> merits, you may want to suggest Warren to implement partial >> renderings, adding startrow, endrow, startcolumn and endcolumn to the >> ray command. That might be handy when using pymol for raytracing >> scriptwise on a cluster. But you'll be disappointed if you expect it >> will keep your scene from crashing Pymol (but then there's the >> hash_setting...). >> >> I hope this clarifies things a bit. >> >> Cheers, >> >> Tsjerk >> >> On Sat, Oct 24, 2009 at 8:21 AM, Seth Harris <set...@gmail.com> wrote: >> > Tsjerk, >> > I think you are missing the point of Tom's post, which was a suggestion >> > to >> > aid someone who's computer could NOT deliver the image in "as high a >> > rsolution as you want', so Tom was proposing a way to break the image >> > down >> > into bite-size chunks that the computer COULD then handle. So the >> > suggestion >> > was to take one scene desired at 9000x6000 which would make the computer >> > crash and instead mathematically figure the necessary transpositions to >> > render it in tiles such as upper left quadrant, upper right, etc. each >> > at >> > 4500x3000 or whatever the computer could handle (and then you could put >> > them >> > all back together in photoshop, e.g.). So for someone with deep coding >> > skills they could likely teach pymol to ray just the segments they >> > wanted >> > and deal with the overall perspective of the whole scene, but for a hack >> > like myself I'd likely start the work around of trying to figure out the >> > camera position and where I'd have to put it to get just a quarter of >> > the >> > scene at a time in the viewport (or an eighth, or whatever). As Warren >> > said, >> > you'd likely have to set orthoscopic to get rid of the perspective for >> > starters (at least in the hack approach, but there would be more direct >> > better ways for the skilled). >> > >> > As a side point, Povray (if I remember correctly) in fact allows you to >> > break a large rendering job up into tiles pretty much just like Tom >> > suggests, so not so crazy of an idea! I think in that case you can >> > specify >> > starting and ending x,y pixel coordinates from your large scene. But not >> > in >> > Pymol as far as I know. >> > >> > Sorry if I've now gone stepping in and misinterpreting someone's points >> > myself, but it seemed like there was a lot of cross-purpose talk either >> > from >> > assumptions or extra politeness! >> > >> > -Seth >> >> >> >> > The requested size is 20" by 30". I calculated that it would >> >> > correspond >> >> > to a 6000 x 9000 pixel image. >> >> > How can I create such a large ray traced image without crashing the >> >> > computer? >> >> >> >> >> >> Message: 2 >> >> Date: Fri, 23 Oct 2009 11:27:47 +0200 >> >> From: Tsjerk Wassenaar <tsje...@gmail.com> >> >> Subject: Re: [PyMOL] Saving high resolution images >> >> To: Thomas Stout <tst...@exelixis.com> >> >> Cc: pymol-users@lists.sourceforge.net >> >> Message-ID: >> >> <8ff898150910230227k12719cedv8656233c7e67d...@mail.gmail.com> >> >> Content-Type: text/plain; charset=ISO-8859-1 >> >> >> >> Hi Thomas, >> >> >> >> You can also zoom out to get everything in view. You can also change >> >> the field of view. And then you can ray just the way you want, based >> >> on what you have in sight, in as high a resolution you want. >> >> >> >> Cheers, >> >> >> >> Tsjerk >> >> >> >> On Thu, Oct 22, 2009 at 9:35 PM, Thomas Stout <tst...@exelixis.com> >> >> wrote: >> >> > >> >> > But isn't it true that only the objects that are visible in the >> >> > viewport >> >> > are what are written to the rendered image file? ?I was proposing >> >> > rendering >> >> > a poster-sized image in "tiles" and stitching them back together post >> >> > facto >> >> > to create a very large, high resolution image. >> >> > >> >> > something like: >> >> > ----------------------- >> >> > | ? ? ? ? ?| ? ? ? ? ?| >> >> > | ? ? ? ? ?| ? ? ? ? ?| >> >> > | render 1 | render 2 | >> >> > | ? ? ? ? ?| ? ? ? ? ?| >> >> > | ? ? ? ? ?| ? ? ? ? ?| >> >> > |---------------------- >> >> > | ? ? ? ? ?| ? ? ? ? ?| >> >> > | ? ? ? ? ?| ? ? ? ? ?| >> >> > | render 3 | render 4 | >> >> > | ? ? ? ? ?| ? ? ? ? ?| >> >> > | ? ? ? ? ?| ? ? ? ? ?| >> >> > ----------------------- >> >> > >> >> > I feel like I'm missing something important here! >> >> > -Tom >> >> > >> >> > -----Original Message----- >> >> > From: Tsjerk Wassenaar [mailto:tsje...@gmail.com] >> >> > Sent: Thursday, October 22, 2009 11:45 AM >> >> > To: Thomas Stout >> >> > Cc: pymol-users@lists.sourceforge.net >> >> > Subject: Re: [PyMOL] Saving high resolution images >> >> > >> >> > Hi Thomas, >> >> > >> >> > The viewport is not important for rendering. You can render at >> >> > whatever >> >> > resolution/dimensions you want with whatever viewport. You can even >> >> > make a >> >> > panorama using a wide angle lens if you want to have something for on >> >> > your >> >> > wall ;) >> >> > >> >> > Cheers, >> >> > >> >> > Tsjerk >> >> > >> >> > On Wed, Oct 21, 2009 at 11:43 PM, Thomas Stout <tst...@exelixis.com> >> >> > wrote: >> >> >> >> >> >> >> >> >> Here's a crazy idea: ?if someone out there were clever at both >> >> >> python >> >> >> and manipulating orientation matrices, I would bet that a "scene" >> >> >> could be quartered or cut into eighths and "translated" such that >> >> >> each >> >> >> portion filled the viewport for rendering; then the individual >> >> >> images >> >> >> could be spliced back together in one's favorite image handling >> >> >> program a la panoramas in photography.... ?Is this way too complex >> >> >> to >> >> >> be bothered with? ?I suspect parallax may be a problem... >> >> >> >> >> >> -Tom >> >> >> >> >> >> >> >> >> >> >> >> -----Original Message----- >> >> >> From: Eva Vanamee [mailto:eva.vana...@mssm.edu] >> >> >> Sent: Monday, October 19, 2009 1:51 PM >> >> >> To: pymol-users@lists.sourceforge.net >> >> >> Subject: [PyMOL] Saving high resolution images >> >> >> >> >> >> Hi, >> >> >> >> >> >> I'd like to save an image in high resolution for a poster. >> >> >> The requested size is 20" by 30". I calculated that it would >> >> >> correspond to a 6000 x 9000 pixel image. >> >> >> How can I create such a large ray traced image without crashing the >> >> >> computer? >> >> >> Many thanks in advance for the help. >> >> >> >> >> >> Best, >> >> >> >> >> >> - Eva >> >> >> >> >> >> >> >> >> >> >> >> ---------------------------------------------------------------------- >> >> >> > >> > >> > ------------------------------------------------------------------------------ >> > Come build with us! The BlackBerry(R) Developer Conference in SF, CA >> > is the only developer event you need to attend this year. Jumpstart your >> > developing skills, take BlackBerry mobile applications to market and >> > stay >> > ahead of the curve. Join us from November 9 - 12, 2009. Register now! >> > http://p.sf.net/sfu/devconference >> > _______________________________________________ >> > PyMOL-users mailing list (PyMOL-users@lists.sourceforge.net) >> > Info Page: https://lists.sourceforge.net/lists/listinfo/pymol-users >> > Archives: http://www.mail-archive.com/pymol-users@lists.sourceforge.net >> > >> >> >> >> -- >> Tsjerk A. Wassenaar, Ph.D. >> Junior UD (post-doc) >> Biomolecular NMR, Bijvoet Center >> Utrecht University >> Padualaan 8 >> 3584 CH Utrecht >> The Netherlands >> P: +31-30-2539931 >> F: +31-30-2537623 > > -- Tsjerk A. Wassenaar, Ph.D. Junior UD (post-doc) Biomolecular NMR, Bijvoet Center Utrecht University Padualaan 8 3584 CH Utrecht The Netherlands P: +31-30-2539931 F: +31-30-2537623 ------------------------------------------------------------------------------ Come build with us! The BlackBerry(R) Developer Conference in SF, CA is the only developer event you need to attend this year. Jumpstart your developing skills, take BlackBerry mobile applications to market and stay ahead of the curve. Join us from November 9 - 12, 2009. Register now! http://p.sf.net/sfu/devconference _______________________________________________ PyMOL-users mailing list (PyMOL-users@lists.sourceforge.net) Info Page: https://lists.sourceforge.net/lists/listinfo/pymol-users Archives: http://www.mail-archive.com/pymol-users@lists.sourceforge.net
