On 10/11/20 2:14 PM, William F Pokorny wrote:
> While thinking of deleting the quilted value pattern, started to toy 
> with the idea of replacing it with a seed pattern having values in the 
> function_interval range of -1 to 1.
> 
...

One more. I had thought I could create another image where I scale the 
seed derived pattern very small creating a really good stress tests for 
the am3 anti-alias mode.

Since that set of cases am1,am2,am3 cases making sure there was no 
overall color shift, I've been carrying this a the am3 mode really worth 
much over heavy am1/am2.

It's valuable here. Fed the input on the left (no AA) - there is  much 
more 'line-ish' detail below what is seen. The am3 mode came out quite a 
bit faster than am2 for equivalent threshold and depth. The fixed am3 
and confidence set at 0.98. Resulting images not identical, but very close.

3 minutes 56 seconds (236.791 seconds)   am3
Pixels: 490000 Samples: 425852553 Smpls/Pxl: 869.09

5 minutes 30 seconds (330.180 seconds)   am2
Pixels: 521284 Samples: 531634923 Smpls/Pxl: 1019.86

869.09 -> 1019.9 ---> 17.35%
236.8s -> 330.2s ---> 39.44%

Bill P.

Post a reply to this message

Attachments:
Download 'am2vsam3.jpg' (131 KB)

Preview of image 'am2vsam3.jpg'