IP BP Ratios - Hurt Iframe Quality?

[HD MOVIE SOURCE]

Hi,

If you go below 1.0, like ipratio=0.90, pbratio=0.90, does this hurt iframe quality?

Or a different question is, does this allow more data to be pushed into p and b frames. And yes it lowers iframe quality, but allows a much higher quality of p and b frames?

Another thought I've had is, does lowering beyond 1 hurt reference frame quality? Or does using P and B frames as reference frames clear up that issue?


My thought here is to shift some of the data hog that an iframe represents and push slightly higher into p and b frames because they offer much high compression. And with x265, p and bframe quality is exceptionally good.

Thanks

[benwaggoner]

Hi,

If you go below 1.0, like ipratio=0.90, pbratio=0.90, does this hurt iframe quality?
It absolutely would if VBV limited. I've never tried values <1.0, and struggle to think of any scenario where that would be beneficial.

Or a different question is, does this allow more data to be pushed into p and b frames. And yes it lowers iframe quality, but allows a much higher quality of p and b frames?
P and B frames will reference the I-frame, so poor quality there gets rippled through the GOP. It's almost always better to optimize quality more the higher up in the reference hierarchy a frame is, as those benefit the most dependent frames. So I>P>B>b. For grainy content having the ratios smaller can help reduce strobing, but the cleaner the content, the bigger default ratios are optimum.

Another thought I've had is, does lowering beyond 1 hurt reference frame quality? Or does using P and B frames as reference frames clear up that issue?
I would not expect it to be helpful. You can test and report your findings.


My thought here is to shift some of the data hog that an iframe represents and push slightly higher into p and b frames because they offer much high compression. And with x265, p and bframe quality is exceptionally good.
Ah! I frames aren't data hogs because they are worse somehow. They are data hogs because they are all intra coded. Intra needs to be coded somewhere! If a reference frame is poorer quality, then frames that reference it need to spend more bits to "repair" that quality, and rarely get as good as if they started with a clean reference. Pushed far enough, you'd wind up with having more intra coded blocks in P and B frames because the I frame was a poor reference, and that's using more bits than getting the I frame correct in the first place.

[PoeBear]

P and B frames will reference the I-frame, so poor quality there gets rippled through the GOP. It's almost always better to optimize quality more the higher up in the reference hierarchy a frame is, as those benefit the most dependent frames. So I>P>B>b

I understand the ranking of the frame types, but how far are the gaps between each ranking, which is more important to focus on? When encoding something, and it gives out QP values in my log, if we're targeting something near transparent, would we want the the P-frames closer in QP value to the I-frames, or the B-frames closer in QP value to the P-frames, or should their numerical QP value gap be equal in both directions? Or that something that would change based on the source-type? Like would grainy movies would be better to have similar gap sizes, and animation would be better to have a larger gap between B and P, and smaller between I and P? Or vice-versa

[benwaggoner]

I understand the ranking of the frame types, but how far are the gaps between each ranking, which is more important to focus on? When encoding something, and it gives out QP values in my log, if we're targeting something near transparent, would we want the the P-frames closer in QP value to the I-frames, or the B-frames closer in QP value to the P-frames, or should their numerical QP value gap be equal in both directions? Or that something that would change based on the source-type? Like would grainy movies would be better to have similar gap sizes, and animation would be better to have a larger gap between B and P, and smaller between I and P? Or vice-versa
The noisier the content, the lower --ipratio and --pbratio should be. Clean animation can be higher, given lots of repeated frames and repeated elements between frames.

Generally --ipratio will be lower than --pbratio.