I teach C++ & computer graphics and make games. Working on a village building game: https://nitter.cf/t.co/skTGGgH0Ay Check out my cozy traffic sim: https://nitter.cf/t.co/7Dzc83p6uq He/him
Joined May 2012
- Tweets39.3K
- Following541
- Followers21.1K
- Likes119K
Pinned Tweet
👀 What's that??
🤯 A new devlog about my village building game!
😅 It only took me 7 months
👇 Link below
Made roulette probability dependent on sample weight & tweaked sun vs sky sampling weights. Less noise now, though still quite a lot. I guess clamping is the way to go 😓
Comparison between scattering coefficients: 5, 20, and 100
A martian-ish cloud
(I really need to get another cloud model...or at least a higher-res version of this one)
One more cosmic-ish multiscatter render with scattering coefficients being kinda inverse of those seen in Earth's atmosphere (red scatters a lot, green scatters somewhat, blue scatters a bit)
Added clamping for the path traced samples, now the image is a bit darker than it should be but at least it converges to something presentable instead of a noisy mess
This is an offline monte carlo volumetric path tracer, using delta tracking to sample the transmission term, unbiased up until the final clamping step. The wavelength-dependent scattering is intentional, it's unrealistic for Earth-like clouds.
Finally implemented multi-scattering in my monte-carlo volumetric path tracer. Kinda not impressed with the result: it's extremely noisy while simply looking much brighter, idk?
Here's a comparison: single-bounce, 2-bounce, 3-bounce, and full unbounded multi-scatter.
Reduced the frequency of the Worley noise by half and now the periodic noise artifacts are basically gone. I think this render is my favourite so far 🥰
Really need to add multi-scattering, should make it look even better
Mixed the disney cloud 3D texture with a higher-frequency Worley noise for some extra detail. I like it but there are some noticeable directional artifacts due to periodic noise...
Volumetric path tracing + very high scattering coefficients + thresholding density make the volume look like a solid surface
Also tried tricubic interpolation of density data. As expected, it's much smoother, removes blocky artefacts, and is generally more suitable for stuff like clouds, at the expense of 8x texture samples.
Left: trilinear, right: tricubic
Ooof, wavelength-dependent scattering + scattered surrounding sky produce incredibly soft & lovely visuals
An attempt at supporting wavelength-dependent absorption & scattering in my volumetric path tracer. Each path simply selects a random color channel to work with 🙃
Some wallpaper content right there
(I'm procrastinating instead of adding multi-scattering lol)
#graphics
Added sampling the sky light in random directions, starting to look so nice
#graphics
Volumetric path tracing adventure continues: added single-scattering by approximating the shadow ray transmittance using ratio tracking. Took a while to get all the coefficients right.
Left: raymarched, right: path-traced
#graphics
