Time v. Quality of AVI -> M2V

[Pratticus]

I have posted this before, but got no answer.

I am curious as to the benefit of performing multiple VBR passes in CCE on an avi source that contains a movie at a relatively low bit-rate.

Specifically, would investing 2-3x more hours of encoding via multiple passes yield noticeable results on a clip that has previously been downsampled from, say 5MBit/s to 1Mb/s?

Thanks in advance.

[bugster]

Its doubtful. There may be some slight improvement but as to wether it is worth the effor, only you can decide. Test on a short clip first to help you.

[SaSi]

The catch is on the way multi-pass vbr works. You decide the average bitrate (should be the bitrate that would create an mpeg stream that is the largest possible but still fits in a DVD leaving space for audio).

CCE will then scan the source stream, calculating the bitrate that provides the best possible quality without wasting bitrate. (this is the first pass. It does not encode but produces statistics).

The second pass takes (from the statistics) the average bitrate estimated (which is higher than the one you selected) and scales it down so that the average bitrate becomes the one you need.

If you select two pass, this pass does the encoding.

Otherwise, this pass recalculates and starts to finetune within GOP regions.

The whole process is repeated for as many passes as you select, with the last pass doing the encoding. This means that all but the last pass are fairly fast.

Tmpgenc does this in two passes. CCE can repeatedly fine-tune.

More than two passes really don't make a difference if you are encoding in a relatively high average bitrate - say 5,000kbps. However, I have found that an average of 3,000kbps (which is close to the SVCD bitrate of 2,600 but with the DVD resolution) can benefit from at least 2 more passes. Although CCE allows you to select any number of passes you desire, I don't think more than 5 passes would make a sense, unless the video has wide variations of high motion and still scenes.

[Pratticus]

Thanks for the replies, bugster and SaSi.

THat gives me 75 percent of what I need to know. I guess my next question would be to what extent can you "clean-up" the source file?

Let's say the source file has minor blocking due to the encoding already done on it. Is it possible to eliminate or reduce that blocking during the re-encoding to .m2v format? Is it possible to essentially reverse some of the bitrate compression?

[bugster]

Thanks for the replies, bugster and SaSi.

THat gives me 75 percent of what I need to know. I guess my next question would be to what extent can you "clean-up" the source file?

Let's say the source file has minor blocking due to the encoding already done on it. Is it possible to eliminate or reduce that blocking during the re-encoding to .m2v format? Is it possible to essentially reverse some of the bitrate compression?

Not really no, the old rule applies, garbage in = garbage out. You can improve things a little bit to your own eyes but its really all about compromises. For instance Tmpgenc has a 'soften block noise' option which helps to remove blockiness but at the expense of picture sharpness. CCE does not have a comparable feature but if our source is mpeg-1 or some form of compressed avi, you could try and find a virtualdub or avisyth filter that will help and then frameserve to CCE. Other than that I can't be of much more help.

[SaSi]

Of course, garbage in = garbage out applies. Theoretically, it is possible to reverse estimage some of the damage blockiness creates. I have seen at least one such filter (jpeg cleanup) in a program called Thumbsplus. It cleans-up jpeg images removing subtle colour imperfections caused by a high compression ratio. Theoretically, such (minor) improvements can be achieved in video - and the algorithm should be similar - however what the algorithm effectivelly does is interpolation, extrapolation and softening. It can cause a subjective improvement, but not much more.

However, let me add yet another method of improving compression. Improving compressibility of the source.

The compression tries to compress and keep as much of the information as possible. If the source video has noise, the mpeg compressor doesn't know it is noise and considers it to be random information it tries to preserve as much as possible. A lot of bitrate is consumed in this futile excercice. Removing such noise from the source can dramatically improve compression with a significanly better picture.

Tmpgenc has an excelent quality filter that is unacceptable in terms of performance. Activating it can multiply compression time by a factor of X2 ~ X3 easily.

VirtualDUB has a less excelent filter that is a lot faster. In certain cases, I use the smoother filter (plug-in filters exist that are better) while clipping and de-interlacing. That can help produce a two hour film mpeg film with a total size of less than 2 Gb with perfectly acceptable quality.

To finish my comments, I would like to point out something I think is a limitation in Tmpgenc. It appears to have problems in dark scenes and rather low bitrates (2600kbps). In such cases, blockiness appears. At high bitrates (>4000kbps)no such thing is observed.