Cosmic Background (Stars and Galaxy) [2019 Night Sky]

Started by WAS, December 24, 2019, 11:59:51 am

Previous topic - Next topic

WAS

Quote from: KyL on December 30, 2019, 12:31:04 pmGreat results here! I am quite convinced by the animation tests, this looks really clean. How many AA were this rendered at? And how fast was it?

I honestly didn't test AA thoroughly. I just did research on AA and found Cubic B-Spline was best for this, and additionally a small pixel sampling which I translated to pixel noise threshold (???). So I used 1/4 samples as Oshyan suggested with pixel noise threshold at 0.005. Render times were pretty high at about an 1hr or less each. AA was at 7. Though take into account my dedicated server is only using half of it's 24 threads effeciently with the Terragen Linux Node. I can't figure out why. So I get about double render time of my Ryzen 5 2600 when it should be comparable.

I will do another test with lower settings.

I'd imagine for atmospheric shots you'd only need to slightly pan/zoom for most scenarios and comp it in, which would mean you wouldn't need as nearly as many frames as the foreground planet shot.

Oshyan

At that low of a pixel noise threshold it may be getting close to "max samples" for practical purposes. I'd be curious how much difference there is between the two actually. Also note that any flicker would be hard to really see with this relatively quick camera move. Flicker is likely to be worst with slower movement. Still it's a great result and does seem to be pretty stable.

- Oshyan

WAS

Quote from: Oshyan on December 30, 2019, 03:10:30 pmAt that low of a pixel noise threshold it may be getting close to "max samples" for practical purposes. I'd be curious how much difference there is between the two actually. Also note that any flicker would be hard to really see with this relatively quick camera move. Flicker is likely to be worst with slower movement. Still it's a great result and does seem to be pretty stable.

- Oshyan

You should be able to slow down the video and examine the frames. And I don't think what you're saying is true, as inherently the largest difference in samples would happen at a greater distance. A slower pan would just allow you to visually see this happening easier. You can see there is twinkle in the denser faint stars, which is the small stars PF, which is only at 220000, which I have increased to 350000 in my next test. Normal stars are around 750000 with a min of 500000 and seem fine.

I did a test at 0.6 MPD and 0.6 AA with 0/16 samples and 0.04 pixel threshold (which was more or less what it set automatically minus the trailing numbers which I trimmed. It was like 0.0436210 or something.). This one took 4 hours and 13 minutes. However, like I suspected, it's not stable, and is a fireworks show.

So unfortunately, it does matter having MPD at 1 and having higher AA with tighter sampling. Ironically it seems stars of the same dimensions are more stable with a colour merged around them (the galaxy glow spread).

WAS

Well, that's something. I just noticed that the sparse stars were at 70000 instead of 700000 too for scale and lead-in, that doesn't help anything.

Jo Kariboo

Excellent works!
Thanks again for your great generosity! :D