Lossless video format for pre-processed videos

[moviefan]

Hi everyone,

I'm currently pre-processing a very noisy video whose picture quality improves a lot by using the RemoveNoiseMC script. The only issue that bothers me is time since this script is so damn slow - the price for quality. I have read, and it makes sense, that encoding a pre-processed video with a lossless codec first and then encoding the whole thing again to say MPEG2 (> 2 passes) spares time as I only once export the video extremely slow applying the filteres and then reencode it a lot faster later on.

But which lossless format should I use for this purpose? A standard uncompressed video of at least an hour needs a vast amount of disc space and somehow exporting the video that way (with VirtualDub) is even slower than encoding it straight to MPEG2 (1 pass) - weird...

How would you proceed in this issue?

Regards

[Zarxrax]

Well for that purpose you will want a codec that's fairly fast, because if you choose a slow lossless codec then you will lose much of the benefit.

I always use Lagarith Lossless codec myself, but the huffyuv and ffv1 codecs in ffdshow are good choices as well. Among these three, ffv1 is slowest but compresses the best, huffyuv is quickest and compresses least, and lagarith is inbetween.

[Trixter]

Seconded. But one advantage Lagarith has over the others is that it's multithreaded, so if you have a dual-core machine, Lagarith will outperform the others.

BTW, you have the right idea; I do 5-pass encodes with my MPEG-2 encoder and I always render to lossless before starting the encode. That way if I have to start over, or I screw up a setting in the encoder, or want to create multiple output targets, etc. I don't have to do all of the preprocessing in avisynth all over again.

[squid_80]

Seconded. But one advantage Lagarith has over the others is that it's multithreaded, so if you have a dual-core machine, Lagarith will outperform the others.
Are you speaking from experience? Because my tests show the multithreading gives little gain, FPS increases by about 0.5.

[Trixter]

On a Core 2 Duo, yes. I'm not sure how you were testing, but you can't do it from a process-heavy avisynth script -- I did a quick verify with a huffyuv file -> lagarith using VirtualDub on Fast Recompress and there is most certainly improvement (at least 30% but I didn't measure exactly) over having multithreading turned off.

[moviefan]

Why do I get the feeling that either HuffYUV somehow encodes in 2 passes or VirtualDub calculates the percental progress of the encoding process wrongly? I have now encoded about 40% of the total frames, but VirtualDub says it has been only 20%. Also the "Total time (estimated)" calculates for exactly twice the time I estimate. Does HuffYUV encode in 2 pass mode or how can you explain this phenomenon?

Edit: By the way, I use ffdshow HuffYUV, if that is of importance...

[foxyshadis]

Definitely not. Virtualdub should put the final frame # on the bottom right, and you might want to be sure that's the number you expect. It's possible to get frame-doubling by forgetting to cut down a bobber's output. If you're using an older version, like vdubmod, it's also possible it's just a bug in vdub.