Swarming Limbs and Twisted Nets: 5th Parade of Hopefuls

June 1st, 2020 by Diana Coman

I publish this early to make sure it’s not published late – the practice of course turned out less orderly than the theory, with some unexpected hiccups at rendering time 1 meaning that I’m writing this while some of the seeds are still being rendered. Rather than force everything to wait for the full set though, I think it makes more sense to publish now a first archive (4.5G) since it contains anyway 16368 pictures (31 seeds * 2 symmetry options * 22 textures * 12 points combinations) and it can take a while to just go through them all, not even considering the couple of hours required to upload/download it anyway. I’ll update afterwards with the screenshots for the remaining seeds and hopefully also with at least one full set of combinations with the more demanding collapsing of points at minimum distance – for which I already have the skeletons and models made but they are waiting in the pipeline for a place on the rendering machine! In hindsight I should have possibly expected that the rendering will still be the bottleneck, no matter that the skeleton generation takes otherwise longer – it’s clearly the graphical part that I’m least prepared for (and keen about…) otherwise and overall, indeed.

Updated 3 June 2020: the archive with remaining seeds, 132 to 150 (“quick collapse”, 2.7G); another archive with the rendered models from one seed done with “slow collapse” for the full set of points+symmetry+textures. (151M).

For the record, here’s the list of changes and updates since last parade:

  1. Fixed the uncovered errors at mirroring.
  2. Choice between “quick collapsing” of points (based on first pair of points found with distance between them less than average) and “slow collapsing” of points (based on picking at every step the pair of points that are closest to one another so at minimum distance). The “c1” in the archive name stands for the quick collapsing (collapse first) while cmin stands for slow collapsing (collapse minimum).
  3. Updated model generation to pick for each bone randomly one of the 2 existing meshes (one is cylinder-based, the other is egg-based) with a 50/50 chance for each. The cylinder shape is also changed to reduce the noisiness and especially the “floating” unconnected bits.
  4. Updated list of textures with 3 textures taken out (2,3, and 7 from the list in the previous parade) and 12 textures added. The current total is 22 textures in use.
  5. Changed the rendering so that the hopefuls are rotated by a different angle around each of the three axes. All three angles are obtained from the MT prng seeded with the texture number so that the rotations are distinct for each texture but the same across different seeds/symmetry options.
  6. Refactored and various changes to the whole model generation chain to streamline it and allow the run of the different parts separately or in parallel where possible. As part and parcel of this, the skeleton generation now spits out also a plain text file, basically “smg format” of the skeleton in addition to the CS/Cal3d format so that any skeletons generated can be more easily reused for any models and/or otherwise further changed, as required. The steps from 0 to rendered hopeful are those: 1. mesh generation (currently only 2 meshes but after this refactoring, it can now easily be extended without requiring any further changes to next steps to use the outputs); 2. skeleton generation; 3. model generation (picking at random from the full list of meshes from 1 and setting them on skeletons from 2; 4. rendering (going through the full list of available models from 4 and the full list of textures, moving files where the client expects them, firing up/killing the client and taking screenshots).

Picked mostly by chance, accident and other happy preferences, the stars of this show include an almost-insect, a twisted arrow-head of sorts and possibly a dog-headed chicken:

skel_101_vol127_127_srf31_31_sym2_tex8_rotxyz35_190_97_1280.jpg

skel_101_vol503_503_srf263_263_sym2_tex4_rotxyz146_342_65_1280.jpg
skel_102_vol47_47_srf23_23_sym2_tex13_rotxyz240_268_92_1280.jpg
skel_104_vol23_23_srf47_47_sym2_tex18_rotxyz272_91_200_1280.jpg
skel_104_vol719_719_srf1319_1319_sym1_tex12_rotxyz318_103_227_1280.jpg
skel_131_vol1319_1319_srf719_719_sym2_tex1_rotxyz297_124_65_1280.jpg

For easy reference, here’s the full list of textures used for this parade:
tex_1_512.jpg
tex_2_512.jpg
tex_3_512.jpg
tex_4_512.jpg
tex_5_512.jpg
tex_6_512.jpg
tex_7_512.jpg
tex_8_512.jpg
tex_9_512.jpg
tex_10_512.jpg
tex_11_512.jpg
tex_12_512.jpg
tex_13_512.jpg
tex_14_512.jpg
tex_15_512.jpg
tex_16_512.jpg
tex_17_512.jpg
tex_18_512.jpg
tex_19_512.jpg
tex_20_512.jpg
tex_21_512.jpg
tex_22_512.jpg

  1. Mostly still that pesky client that apparently is not that orderly within the range of meshes per second measured previously, ugh.[]