Thursday, February 28, 2013

Screen Mode Comparisons

Same as before, just with appropriate screen mode for the stars.  All images color corrected (levels- midpoint to 1.37, saturation to +40).

Low Res: 3.3M splats, 2.38min
Med Res: 7.6M splats, 5.72min
Hi Res: 9.7M splats, 7.13min

Wednesday, February 27, 2013

Screen Mode Low Res

I accidentally used normal compositing mode for the stars instead of screen mode for testing.  Here are frames 0 and 400 in the low res set and their respective parameters.

Exposures: .0008, .001, .0002; Size: 250%; Core: 1.0
Exposures: .005, .003, .001; Size: 125%, 125%, 100%; Core: 0.35
Frame 400 with color correction (to prove we can get close to the look from before).

Low, Med, Hi Res Comparisons

Quick comparison of render times, particle counts, and quality.

Frame 200:

3.3M splats, 2min
10.4M splats, 4.97min

7.7M splats, 2.24min
Frame 400:

3.3M splats, 2min.
7.7M splats, 5.47min
10.4M splats, 7.58min.
Frame 500:

3.3M splats, 4.08min
7.7M splats, 9.33min
10.4 splats, 12.35min

Low-Res, Frames 0 - 400

In testing the first 400 frames, I found an issue with render artifacts.  These stem from the falloff ramps of each splat in the set.

The solution seems to be to render the splats slightly larger and with no falloff for the first frames, and interpolate to the correct size and trapezoid splat type by frame 400.  Here are frames 0, 100, 200, 300, 400.

Trapezoid core: 1.0, Size: 200%
Trapezoid core: 0.675, Size: 175%
Trapezoid core: 0.35, Size: 150%

Trapezoid core: 0.35, Size: 125%
Trapezoid core: 0.35, Size: 125% (dust, nebulae), 100% (stars)

Tuesday, February 26, 2013

Low Res Set: First Images

First set of test images from the low res set (3.3M splats).  Looks pretty good in terms of color, balancing the LOD will be a pain.  Using the square_root splat type

Frame 0.  15sec. NOTE: render artifacts are ugly...
Frame 925. 13min, 1.5M + 23M splats = 24.5M

Low, Medium, and High Resolution Galaxy Sets

For testing purposes, I've created 3 galaxy sets of different resolutions.  I don't know if I'll need the high resolution set for the final renders, but I thought it would be best to have it ready to go.

Low Set: 3,238,933  3,211,856
Nebula: 33,011
Dust: 453,181   426,104 (now varies in height with stars)
Stars: 2,752,761

Medium Set: 7,727,653  7,663,880
Nebula: 149,806
Dust: 1,443,406  989,736 (see above)
Stars: 6,524,338

High Set: 10,538,544  10,390,792
Nebula: 266,601
Dust: 2,433,855  2,285,603 (see above)
Stars: 7,838,588

Stars with Thickness Variation

I decided to vary the thickness of the stars to create a bulge shape in the center.  Also, the outer edges of the galaxy tend to be thicker due to disruption by nearby galaxies' gravitational forces.  This variation helps liven up the edge on view of the galaxy a bit.

Sunday, February 24, 2013

M51 LOD Comparison

I did a quick test to see the difference between the images produced with the LOD code and without.

Without LOD, 20min render, 4.6M particles

With LOD, 12min render, 4.6M initial, 18M LOD (yikes!)

M51 3 Layers, w/ Nebula Variation

Starting to look better! I'm going to try to add some more thickness variation in the stars layer and perhaps a little noise to break up the even distribution.

All 3 Layers, Nebula Thickness Variation from Kendall Litaker on Vimeo.

Saturday, February 23, 2013

Nebula Layer and LOD

Test with nebula layer splats.  Original set of 120K, with 100K - 280K LOD splats generated.  Roughly 10-20s renders.

Nebula Layer Thickness

The nebula layer needs to have more variation in its thickness.  Specifically, each nebulae is roughly as thick as it is wide.  I decided to use a ppm image to roughly delineate thick and thin areas of the layer.

Each color has a specific gaussian PDF function associated with it, thus clamping the thickness to a specified range of values.  Here are the results in just the nebula layer.

In combination with the other 2 layers...

gas: 149475, dark: 884284, stars: 6273772, 3.28 min

With gamma and saturation corrections...

Friday, February 22, 2013

M51 LOD Testing

Testing the LOD code with the M51 data and a section of the most recent camera move.

One thing I noticed was that using gaussian splats removes a lot of the speckly feel I was getting with the trapezoidal splats. 

4835232 Trapezoidal Splats, no LOD, 2.5 min
4835232 Square Root Splats, no LOD, 2.63 min
4835232 Gaussian Splats, 2782650 LOD Splats, 10.82 min
The issue is that gaussian splat type right now adds an extreme amount of LOD spawned splats to the render set, causing huge render time increases.  Not sure if this is supposed to happen or if its a bug in the program. 

Also, closer to the galaxy, the stars start to blow out despite having the same exposure as farther away.  This is without any extra LOD particles, so its just the fact that the camera is closer to the data set.  We'll need some way of animating the exposure over multiple frames.

Trapezoidal Splats, no LOD, exp= 0.005
Trapezoidal Splats, no LOD, exp = 0.001

Final Camera Move, V2

Still needs a few more tweaks, but getting closer...

Camera Move, V2 from Kendall Litaker on Vimeo.

Wednesday, February 20, 2013

Gaussian PDF Added for M51

I felt that the quadratic PDF was creating too harsh of a falloff on the galaxy layers, so I've added a gaussian version.  It seems much smoother.

quadratic PDF
gaussian PDF

M51: All 3 Layers

Combined all 3 layers of M51 again to look at where the dust was at in terms of lookdev.  It's looking much better.  I'll be posting a clip of the tilt cam when I have it.

Straight from render code.
Gamma and saturation tweaks.

Level of Detail Control

Dr. Summers has come through with a level of detail control for the C code splat renderer.  As particles get closer to the camera, the code subdivides them and cross fades between the 2 states.  Still needs testing for the gigantic set of particles we will have, but the initial test looks good.  Initial set is 1000, then subdivides to a max of about 50K.

Level of Detail Control for Splat Renderer from Kendall Litaker on Vimeo.

M51 Dust Noise

In order to capture a little bit of the variety seen in the dust edge on, I've added a noise based cull and shift to the dust particles.  A larger sized noise cuts into the particle set based on the distance from the center creating a more varied edge.  The second noise will shift the positions of the particles in the Y axis to create even more variety.  Below is a close up of the dust- I'm not sure we'll be able to get that detailed, but we can get closer using noise methods.

Clip of the noise with the large scale noise applied. (Stars are black and used to create a mask so we can see the noise correctly.)

Dust w/ Noise from Kendall Litaker on Vimeo.

Saturday, February 16, 2013

Dust + Stars + Nebula

Okay, so here's the first attempt at combining all 3 layers together.  Some obvious issues with balancing the 3 to get it to look right. 

Stars- maybe just need a whole bunch more... they seem to be overwhelmed by the darker dust/nebula particles.

Dust- Current alpha may include too much of the faint dust areas, creating a bit of a pileup in areas.  May need to tweak original alpha and recull to get a more spidery feel.  Also, a fractal noise added to the density or alpha of the dust in the vertical direction may help the spidery/tendrilly feel in the original Hubble image, particularly when viewed from the side.

Nebula- Doesn't integrate as nicely as I would like.  Too dark around edges?  Also no real hotspots in center where star formation is taking place.  This was removed in the original texture map, but it may need to be added back in.  Also, nebulae don't naturally extend throughout the height of the galaxy, they're usually about as deep as they are wide, so fixing that should probably be on the list too.

Total: 6,647,621

Stars: 4,781,374
exposure: 0.02
radius: 0.05

Dust: 1,468,095
exposure: 0.005
radius: 0.05 - 0.15

Nebula: 398,151
exposure: 0.005
radius: 0.05 - 0.15

Dust + Stars w/ Screen Mode

Combined the dust and stars together to play with exposure levels.  With the screen mode, the stars seem to start washing out in terms of saturation- hopefully this can be fixed in post.  With the addition of the dust, the stars exposure level could be picked up a bit.  Its going to be tricky balancing the 2 so as not to get a dirty looking or overly bright galaxy.

Stars: 4,781,374
Exposure: 0.01
Radius: 0.05

Dust: 1,468,096
Exposure: 0.0025
Radius: 0.05 - 0.15