I've been having a problem in quite a few scenes where I use both SSS and bidirectional tracing: the scene starts out fine, but eventually (sometimes quickly, sometimes after a while) new pixel groups start appearing which are extremely bright, and end up overwhelming the image. Even if I run for a long time further, they don't average away. If I bring up the floating-point image in violet, I find that the image is so bright and uneven that nothing can be done to salvage it.
I usually end up turning bidir off when this happens, but I have had some scenes where I want to do both SSS and caustics, and the caustics fill in siginificnatly slower with bidir off. The bright-pixels problem seems to independent of the metropolis and hybrid settings.
I'm attaching some recent SSS test grids which show the problem,and the igs for the first one (for the curious: no modeller, just hand-edited igs and a perl script to generate the xml for the sphere grid).
I've been seeing this problem on all the versions since I first found indigo (since the last 07 version). The attached pix are with 09_test3. I haven't been able to find any posts on this issue (though I've never been able to look at the known-issues list), so I wonder if it's something about my system (Win XP pro x64, 2 xeon x5355s (quad cores), 8 gb ram). I haven't thought of trying it on an x32 machine (until I just typed this...)
Also, I'm surprised to see that, in some of the spheres near the right-hand ends of the rows (high scattering), whole quadrants of the spheres appear not to have been touched by the SSS. I'm also attaching a non-bidir pic I did in 08_stable, when I first was dong these grids, which appears to show this artifact in a non-bidir case, where the bright-pixel problem didn't occur, so I think these two issues may not be related.
I could't attach the inifile, so I'll copy it here:
------Output image dimensions------
"width" "640" --Note that the ratio of these should match the camera aspect ratio defined
"height" "480" -- in the scene xml file.
------General tracing parameters------
"max_depth" "10000" --max ray bounce depth.
"bidirectional" "true" --Use bidirectional path tracing / bidirectional MLT?
"metropolis" "true" --if true, use MLT, if false, use pathtracer
"hybrid" "true" --If enabled, direct lighting is computed with path tracing.
-- Indirect lighting is computed with MLT.
------Metropolis light transport settings------
"large_mutation_prob" "0.4" --Probability of using fresh random numbers.
"max_change" "0.01" --max mutation size, must be in range [min_change, 1)
"max_num_consec_rejections" "1000"
------Network rendering settings-----
"frame_upload_period" "30.0" --number of seconds between uploads of image to network master
------Miscellaneous settings-----
"halt_time" "-1.0" --time in seconds after which to halt the program. Halting is disabled if negative.
"halt_samples_per_pixel" "-1.0" --number of samples per pixel after which to halt the program. Halting is disabled if negative.
"logging" "true" --write to log.txt?
"display_period" "15.0" --number of seconds between display of the render in preview window.
"image_save_period" "60.0" --number of seconds between saves of the render to PNG, EXR and IGI.
"save_tonemapped_exr" "false" --Write .exr tonemapped render as well as normal PNG?
"save_untonemapped_exr" "false" --Write .exr untonemapped render as well as normal PNG?
"save_igi" "true" --Write Render to Indigo Image (.igi)?
"num_threads" "6" --Number of render threads to use. Set to number of cores for fast rendering.
"cache_trees" "true" --Cache kd-trees on disk?
"polarisation" "false" --broken, leave it off.
"super_sample_factor" "3" --a higher value takes more RAM but reduces aliasing along high contrast edges.
"bih_tri_threshold" "50000" --Any meshes with at least this number of triangles will use a Bounding Interval
--Hierarchy (BIH) acceleration structure instead of a kd-tree
"ray_origin_nudge_distance" "0.0001" --don't fuck with this if you don't need to.
Problem with SSS and Bidir
Problem with SSS and Bidir
- Attachments
-
- the 2nd test rendered in 08_stable, with bidir off (I forget how many spp). Note how some of the high-scattering spheres seem to have quadrants of sss missing.
- im1183768071.jpg (68.52 KiB) Viewed 1476 times
-
- spherestest04.igs
- the scene file for the first test (along with the other attached file testshell01.igs)
- (57.48 KiB) Downloaded 148 times
-
- testshell01.igs
- (2.27 KiB) Downloaded 150 times
-
- 2nd test grid with bidir off, brightened
- sss_problem2_nobi_met_hy_brightened.jpg (93.75 KiB) Viewed 1476 times
-
- test 2, bidir on. This grid compares scattering (horiz) against absorbtion (vert).
- sss_problem2_bi_met_hy.jpg (82.93 KiB) Viewed 1476 times
-
- test 1 with bidir off (met still on). Interestingly, I had to brighten the image compared to when I rendered it in indigo_08_stable. I wonder if it's related to the new <gamma> requirement in the rgb fields (I used 2.2).
- sss_problem_nobi_met_hy_brightened.jpg (92.09 KiB) Viewed 1476 times
-
- test 1: sss test grid, bidir on. horizontal: scattering coef increaing to the right; vertical, ior increasing upwards. All these tests about 1000 samples/pixel
- sss_problem_bi_met_hy.jpg (76.09 KiB) Viewed 1476 times
Who is online
Users browsing this forum: No registered users and 4 guests