3 pass encoding, the theory

[karasu]

I understand why a 2 pass encoding would be usefull for a high quality video with predicted file size.
What's about a 3 pass encoding. What's the meaning of each passes?
I've not found informations elsewhere.

[Sagittaire]

Npass is simply RC refinement. In theory Multipass objective is to make constant quality (for eyes) and with predicted file size.

Prediction between first and second pass can be hard (for example first pass at quant 2 with XviD and second pass at average quant ~q8). In this case use third pass can done better result. IMO multipass is usefull simply for low quality encoding (high average quantizer) or for very particular RC constraint like vbv.

[omion]

The idea is that even in 2-pass mode the encoder can't hit a perfectly consistent quality. The third pass is done to rearrange the bits between the frames.

Each additional pass gets the resultant video closer to the "optimal" quality (defined by the encoder)

The only time a third pass has made a difference for me is when the first and second passes are of very different bitrates. The second pass would have a hard time guessing the proper bitrate for every frame, so another pass would be beneficial. Otherwise I wouldn't bother.

[edit: oops. too slow ;)]

[foxyshadis]

This is why megui sets the first pass to your second pass bitrate, and Teeg recommends setting very specific quants for xvid first pass. In theory 3pass can lay out GOPs more optimially, but in practice xvid and x264 are already very smart about looking some ways ahead to determine effective GOP boundaries.

In general, the more useful multipass is, the more broken the rate control is. (And vice versa.) Divx suffered from bad rate control for a long time, as did several DVD encoders, so they introduced multipass as a cheap workaround.

[karasu]

Ok, so with x264 the gain in quality is marginal for the extra encoding time. Thanks for the informations.