This article was last updated on August 16th, 2009.
Codec Wars 2009
Lossless and virtually lossless codecs
Inspired by Graeme Nattress's review of BitJazz's SheerVideo from 2004, I downloaded the demo version of SheerVideo and did some tests. I have compared a few popular lossless and lossy codecs.
I rendered out the 150 frames Motion 4 HD Template Crowd Control, in 16-bit best quality TIFF at 1280x720 with 29,97 frames per second. That TIFF clip was transcoded with QuickTime Player and each transcode was timed manually.
After the transcoding was done, I imported the clips into Final Cut Pro 7, where I made a new sequence and added the TIFF clip. Then I put each clip above on V2 and changed the composite mode to Difference. For each clip, I exported a TIFF still image, opened them in Preview and turned the brightness to full to see the difference clearly. Note: For the ProRes 4444 clip, I had to push the gamma up to 5 with the levels filter on an adjustment layer i After Effects to be able to see the difference (Thanks Graeme!).
Test hardware and software
- MacBook Pro 2.16GHz Core Duo
- 2GB ram
- 320GB 7200rpm hard drive
- Mac OS X 10.5.7
- QuickTime 7.6.2
- Final Cut Studio 3
- After Effects CS3
- SheerVideo HD Pro for Mac Intel 2.7.7.2
- Cineform - NEO HD Mac v2.3.15 Build 101
- Avid Mac Codecs LE 2.0
Transcoding results
| Codecs | Time | File size | Compression |
|---|---|---|---|
| ProRes 4444 | 24 seconds | 105,9 MB | Lossy |
| BitJazz SheerVideo RGB(A) 10bf | 25 seconds | 162,6 MB | Lossless |
| Animation @ 100% (with keyframes set to all) | 44 seconds | 361,8 MB | Lossless |
| Photo JPEG @ 100% | 22 seconds | 51,1 MB | Lossy |
| Cineform - NeoHD RGB(A) 4:4:4 encoding | 21 seconds | 43,1 MB | Lossy |
| Avid DNxHD 10-bit RGB color UC alpha | 40 seconds | 230,4 MB | Lossy |
| ProRes HQ with 4:4:4 chroma filtering | 33 seconds | 69,3 MB | Lossy |
I am so far pretty impressed with Apple's new ProRes 4444. It looks great, it is fast, and leaves the smallest footprint behind compared to BitJazz and Animation. ProRes 4444 is lossy though, which BitJazz and Animation are not. It remains to see how well ProRes 4444 handles color space conversions. Apple are well known for their gamma issues in QuickTime.
Animation and Photo JPEG are still very good codecs, and I will continue to use them where fit. Photo JPEG is super efficient and fast (and at 100% it is 4:4:4), but it does not support alpha channels and is one of the most lossy of the codecs tested. On the downside, they are both 8-bit codecs. All the other codecs in this test are 10-bit, except for ProRes 4444 which is 12-bit.
Cineform's NeoHD codec is even faster than Photo JPEG, and the file size is smaller. It is a 10-bit codec (Neo 4K is 12-bit) with a lot of interesting features and workflow solutions. One thing to note, is that Cineform claims that QuickTime Player is an 8-bit (only) application.
"NOTE: We don't recommend rendering out of QT Player. The reason is that QT Player is an 8-bit (only) application. If you do choose to render from QT Player, make sure you change your 8-bit DeBayer algorithm to "CF Advanced 1" (or other) algorithm before you begin the render, otherwise it will use the Bilinear DeBayer which is not high enough quality for renders."
If this is true, it might just render this whole test obsolete. Wether it is true or not, I do not know. I thought QuickTime Player just was a front end to the QuickTime export components. I am certainly no codec expert, but I find the topic and results from this test interesting.
BitJazz's SheerVideo sure is lossless, and it is half the file size compared to Animation. Very impressive! Their Synchromy technology also sounds pretty amazing – "SheerVideo incorporates Synchromy nondestructive color-conversion technology —the most accurate color-space converter in the world." Tough to compete with that.
There have been very little news and discussion about BitJazz lately, but that might be about to change soon. Marco Solorio, owner of OneRiver Media, claims something big is coming soon. In a response to Bob Flood's question at Creative Cow:
"I am wondering if ProRez has made the Sheer codecs Obsolete? Have they stopped development?"
(Apr 14, 2009 at 5:39:38 pm)
Marco replied:
"I'm surprised they're not responding, maybe busy with pre-NAB stuff. But what I can say is... stay tuned for their next developments. Three words... Holy. Freaking. Cow." (Apr 14, 2009 at 9:14:46 pm)
Avid DNxHD is still a 4:2:2 codec. The large file size reported in this test is mainly because of the uncompressed alpha channel. With a compressed alpha channel, the file size is 131,3MB. Avid is yet to come with a 4444 version of DNxHD, but as Brian Williams wrote in a comment on Philip Hodgetts' blog post, Why I was wrong about ProRes:
"Avid is more likely to create a DNxHD 4444 codec than Apple is to take ProRes cross platform."
(August 4th, 2009 at 4:34 pm)
All codecs in this test are visually lossless for several generations, but ProRes 4444 is the only codec which supports Apple's Real-Time Extreme engine, which is an important advantage in a Final Cut Pro environment. Will we ever see RT Extreme support for BitJazz? That would be awesome.
Competition is good. I really like the fact that we get to choose between so many great codecs, but deep-down, I really wish the camera and post industry could just collaborate and agree on one UNIVERSAL codec.
Difference test
Source TIFF (frame 52)
ProRes 4444
Had to push the gamma up to 5 with the levels filter on an adjustment layer i After Effects to see the difference.
BitJazz SheerVideo RGB(A) 10bf
Obviously, with lossless codecs like SheerVideo, you can not see any difference.
Animation @ 100% (with keyframes set to all)
The same goes for Animation.
Photo JPEG @ 100%
Cineform - NeoHD RGB(A) 4:4:4 encoding
Avid DNxHD 10-bit RGB color UC alpha
ProRes HQ with 4:4:4 chroma filtering
You can easily do this test yourself. If you want to compare your results with mine, just use the Motion 4 HD Template Crowd Control as your source.
For discussions, please join in at RedUser.net
Twitter: RetweetIf you have any questions or comments, please feel free to contact me:
Twitter: @thomasberglundE-mail: thomas.berglund (AT) gmail.com