SMAA weights not working as expected #663
-
Hey there! I'm currently trying to improve the antialiasing of a webgl Three.js app. We are using the To me, it looks like my edge detection is fine, but the weighted result that I get are "broken". If I compare my results to what I see in the interactive demo, then it feels like I'm doing something wrong. Do you have an idea what I could be missing? Or is this maybe even expected, since my scene is quite dark and my edges are relatively small. screencast.mp4 |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
Hi, the weights seem to be calculated correctly and the detected edges seem to be smoothed as expected. If you're talking about the broken lines and shimmering in motion, then that's not something SMAA can fix. You'd need to use MSAA to get more stable highlights in a dark scene like this. TAA would likely also yield better results in this case, but that's currently not included in For SMAA I'd recommend sticking to the color edge detection with depth-based predication. Use a low |
Beta Was this translation helpful? Give feedback.
Hi, the weights seem to be calculated correctly and the detected edges seem to be smoothed as expected.
If you're talking about the broken lines and shimmering in motion, then that's not something SMAA can fix. You'd need to use MSAA to get more stable highlights in a dark scene like this. TAA would likely also yield better results in this case, but that's currently not included in
postprocessing
.For SMAA I'd recommend sticking to the color edge detection with depth-based predication. Use a low
predicationThreshold
around0.0002
, apredicationStrength
of1
and apredicationScale
of2
. Also, make sure to optimize your depth buffer range by setting the camera'snear
plane to something like0…