mimungr
11th July 2005, 21:41
I did some constant quality encodes with both DivX 5.2.1 and DivX 6.
Settings were: Quantizer: 2; Quantization type: H.263; Adaptive Single Consecutive B-frames; QPel, GMC, and PVE all disabled. For DivX 5, I used Slow and for DivX6, Better.
I tried two movies. For one, DivX 6 was 23% bigger and for the other 14%. I looked at them in MPEG-4 Modifier. DivX 6 used slightly more I and P frames, but not enough to account for such a big size difference. The perceived quality was about the same -- they all looked really good (no surprise at Q2).
I computed the SSIM for the movie that grew by 23%, and DivX 5 actually had a slightly better score despite the smaller size.
Since both encoders are using the same quantization, the only explanation I can think of is that DivX 6 is doing a poorer job of predicting blocks. But it doesn't seem likely that it would regress to this extent.
So I must be doing something wrong, but I can't figure out what.
Any ideas?
Settings were: Quantizer: 2; Quantization type: H.263; Adaptive Single Consecutive B-frames; QPel, GMC, and PVE all disabled. For DivX 5, I used Slow and for DivX6, Better.
I tried two movies. For one, DivX 6 was 23% bigger and for the other 14%. I looked at them in MPEG-4 Modifier. DivX 6 used slightly more I and P frames, but not enough to account for such a big size difference. The perceived quality was about the same -- they all looked really good (no surprise at Q2).
I computed the SSIM for the movie that grew by 23%, and DivX 5 actually had a slightly better score despite the smaller size.
Since both encoders are using the same quantization, the only explanation I can think of is that DivX 6 is doing a poorer job of predicting blocks. But it doesn't seem likely that it would regress to this extent.
So I must be doing something wrong, but I can't figure out what.
Any ideas?