MP4 encoding questions

[jagabo]

If you want to compare results, here's a very clean 140 MB 720x480 24 fps HuffYUV AVI file as starting material:

http://www.mediafire.com/?27x992ir1bnry3d

Attached is my veryfast CRF 18 encode with x264

x264.exe --preset=veryfast --crf 18 --ref 2 --bframes 2 --sar 32:27 --output %1.mkv %1

A handbrake encode with the settings I gave earlier looks about the same.

[smitbret]

Well, I think I'm converted. I ran some encodes using Handbrake and an approximation of Jagabo's settings to re-encode the first 31 minutes of The Mummy. The CPU Veryfast preset was definitely better. I ran through each test twice to be sure. Some blocking in both, especially in dark clothing and high action. The artifacts in the CUDA encode were just overwhelming. The CPU encode was about 5-7% faster with my Phenom II x4 955.

I doubt that I'll ever use the Veryfast settings, but I'll tweak the Trellis, Motion Estimation Method, Subpixel ME, 4 ref frames and see where that leads me. I may run a comparison on some higher bitrates and see what happens.

Incidentally, I used the 'remove pulldown' feature when I demuxed the original .vob file using tsMuxeR rather than the detelecine feature during the re-encode and it worked much better.

[jagabo]

I doubt that I'll ever use the Veryfast settings, but I'll tweak the Trellis, Motion Estimation Method, Subpixel ME, 4 ref frames and see where that leads me. I may run a comparison on some higher bitrates and see what happens.

Stepping up from (veryfast) subme 2 to 3 smooths out most of the roughness in motion areas (and usually increases the file size by 5 or 10 percent) but takes about 25 percent longer to encode. It might be worth the extra time and disk space if you're picky.

Regarding reframes: if you flip back and forth between 2 and 4 reframes images you can see difference but it's hard to say which is better. The difference in file size is usually very small (~1 percent). But the difference in encoding time is also very small so using 4 reframes won't cost you much.