BitJazz: Products: SheerVideo: Speed: Prosumer-Quality Codecs

BitJazz
SheerVideo Speed
Prosumer-Quality Codecs

What if broadcast quality is unaffordable and all our live sources are in DV? Given the low quality of the DV sources, is there any point to using perfect-fidelity SheerVideo for DV work?

The main factors that traditionally make studio-quality and broadcast-quality video so expensive are

the camera
the recording decks
the video capture card
RAID arrays

If your video source is DV, you've already skipped the first three of these wallet-busters, whether you continue to recompress in DV and accept the generational loss, or choose to work in nondestructive SheerVideo during production. Both DV and SheerVideo are fast enough that they don't need RAID storage, and can make do with ordinary disks instead, although both can benefit from fast RAID when dealing with multiple video streams.

Editing in SheerVideo has three advantages over DV:

SheerVideo compresses way faster than DV
SheerVideo always displays with perfect fidelity, whereas the QuickTime DV codec displays in low quality by default
SheerVideo always perfectly preserves the image, whereas DV further degrades the image with each generation of decompression and recompression

However, editing in SheerVideo also has one disadvantage compared to DV:

SheerVideo takes more disk space than DV

So in the case of DV, there's a tradeoff. Where speed, viewing quality, and image preservation are more important to you, use fast lossless SheerVideo. Where disk space overrides these concerns, use DV.

To see just how much faster SheerVideo compresses than DV, here's a chart comparing the RAM-to-RAM speed performance of QuickTime DV versus Sheer, as measured with Apple's microsecond timer on the Kodak Test Set. Because the QuickTime DV codec only supports SD resolution, the images for this test were scaled down to 720x480 pixels using Adobe Photoshop's bicubic interpolation.

Note that the DV codec does not support Y'CbCr 8bv 4:4:4 formats.

Codec	Format	b/S Error	Power	MiB/s Speed	= encoding = decoding

SD Video	RGB 8b	-	-	29.7	reference speed
DV	RGB 8b	2.48 Medium	9.49	19.1 54.0
Sheer	RGB 8b	0.00 Perfect	2.20	110.2 100.9

SD Video	Y'CbCr 8bv 4:2:2	-	-	19.8	reference speed
DV	Y'CbCr 8bv 4:2:2	3.75 Medium	6.33	14.3 87.6
Sheer	Y'CbCr 8bv 4:2:2	0.00 Perfect	2.43	113.1 88.7

*Prosumer-Quality Video Encoding & Decoding Speeds on a 1GHz G4*

As you can see SheerVideo compresses Y'CbCr 8bv 4:2:2 images more than 7 times as fast as DV, and compresses RGB 8b images more than 5 times as fast as DV. In decompression, SheerVideo is slightly faster than DV for Y'CbCr 8bv 4:2:2, and nearly twice as fast as DV for RGB 8b.

So what?

To begin with, a 1 GHz G4 isn't fast enough to compress DV in real time, even excluding disk access. Yet a 1 GHz G4 is fast enough to compress SheerVideo 5 times as fast as real time. Even if you already bought a more powerful computer just to be able to do DV in real time, the overhead involved in compressing the DV output stream is so high that there's little time left to do any editing, so only the most rudimentary editing tasks can be done in real time. SheerVideo is so much faster than real time that it leaves plenty of time over for complex editing tasks.

Note that DV only has one quality setting, because the video stream is always exactly 25 Mib/s. The compression powers given here for DV are just for the video track, and ignore the DV audio track, which always takes up space even if it's empty.

Note that the error rate for DV on this real-world test data is astonishingly high: 3.75 bits per sample. Considering that the worst error rate possible for DV's nominal 8-bit samples is 7 bits per sample (corresponding to completely random guesses), this means that in practice only 4 bits per sample in DV are meaningful information. Contrast this with SheerVideo's zero error rate, preserving the full 8 bits of information per sample.