[FFmpeg-devel] FFv1.3: multipass

Peter B. pb at das-werkstatt.com
Sat Nov 3 18:49:33 CET 2012


On 09/24/2012 02:12 PM, Michael Niedermayer wrote:
> Another test that could be run is testing several short individual
> files individually encoded with 2pass mode vs them concatenated and
> encoded with 2pass mode
> how much worse is it if they are concatenated. This would tell me
> if 2 pass mode needs to be worked on or how much could be gained from
> such work
I've ran the first multipass tests. The results are impressive! :)

I've taken 10 minutes from a VHS source material and 2pass encoded it
(a) in one piece and (b) cut in 10-second segments.

The FFv1.1 source is ~3.6 GiB:

3877389548 input/mthk/amerika_intro-10min-ffv1-VHS.avi

The FFv1.3 files were all encoded with coder/context=1, GOP=1 - and once
with 4 and 30 slices.
Here are the FFv1.3 files, where "1p=1st pass" and "2p=2nd pass"

3712758228
amerika_intro-10min-ffv1-VHS-ffmpeg_3l_1cn_1c_001g_08t_04s_1crc_1p.avi
3496244176
amerika_intro-10min-ffv1-VHS-ffmpeg_3l_1cn_1c_001g_08t_04s_1crc_2p.avi
4019903890
amerika_intro-10min-ffv1-VHS-ffmpeg_3l_1cn_1c_001g_08t_30s_1crc_1p.avi
3550561232
amerika_intro-10min-ffv1-VHS-ffmpeg_3l_1cn_1c_001g_08t_30s_1crc_2p.avi

As you can see, the size difference for this 10 minute sequence is about
206 MiB for 4 slices and 447 for 30 slices.

I have to leave now, but I'll post the rest of the results ASAP.

Regards,
Pb



More information about the ffmpeg-devel mailing list