Doom9's Forum - View Single Post - RV9 2-pass VBR improvements

karl_lillevold · 1st November 2003, 01:24

thanks for trying the new VBR!

Let me illustrate the 2-pass advantage with some numbers from my 3 min test clip at 750 kbps (2min 30sec low action, 30 sec high action)

Code:

M7           Avg     Min
1-pass VBR : 45.81   32.69
2-pass VBR : 45.85   34.25

New
1-pass VBR : 45.86   32.85
2-pass VBR : 46.06   36.51

As you can see:
1) Min PSNR increased by 2.26 dB in new version 2-pass VBR, relative to 2-pass VBR in M7.
2) Min PSNR increased by 3.66 dB from 1-pass to 2-pass in new version, and 1.56 dB in M7, even though Avg PSNR is about the same in 1-pass vs 2-pass
3) Avg PSNR increased 0.21 dB in new version 2-pass VBR, relative to 2-pass VBR in M7

The reason for this behavior is that in 1-pass, the encoder does not know what lies ahead, and starts out the high action at very high quality. Then as it runs out of bits, it increases the QP dramatically. On average, PSNR does not suffer too greatly, but the minimum PSNR gets really bad at the end of the high action scene. As you have seen, the average can in fact sometimes be better for a 1-pass VBR encode, than a 2-pass VBR encode, but the quality is most likely much more variable.

1st November 2003, 01:24	#9 \| Link
karl_lillevold Moderator Join Date: Oct 2002 Location: Seattle Posts: 1,584	thanks for trying the new VBR! Let me illustrate the 2-pass advantage with some numbers from my 3 min test clip at 750 kbps (2min 30sec low action, 30 sec high action) Code: M7 Avg Min 1-pass VBR : 45.81 32.69 2-pass VBR : 45.85 34.25 New 1-pass VBR : 45.86 32.85 2-pass VBR : 46.06 36.51 As you can see: 1) Min PSNR increased by 2.26 dB in new version 2-pass VBR, relative to 2-pass VBR in M7. 2) Min PSNR increased by 3.66 dB from 1-pass to 2-pass in new version, and 1.56 dB in M7, even though Avg PSNR is about the same in 1-pass vs 2-pass 3) Avg PSNR increased 0.21 dB in new version 2-pass VBR, relative to 2-pass VBR in M7 The reason for this behavior is that in 1-pass, the encoder does not know what lies ahead, and starts out the high action at very high quality. Then as it runs out of bits, it increases the QP dramatically. On average, PSNR does not suffer too greatly, but the minimum PSNR gets really bad at the end of the high action scene. As you have seen, the average can in fact sometimes be better for a 1-pass VBR encode, than a 2-pass VBR encode, but the quality is most likely much more variable. __________________ This information is provided "AS IS" with no warranties, grants no rights, and reflects my personal opinion. Last edited by karl_lillevold; 1st November 2003 at 01:27.