FS7: prores, xavx-i and photo Jpeg

[jcs]

To let the audience know that: Cineform gives the same quality as ProRes with half the bitrate, which apparently you spread around on this forum as some kind of mantra, is absurd.

I did some tests with ProRes and Cineform for a very highly detailed locked off camera shot (the beach with buildings in the background). Whereas in a prior test with lots of background blur where I observed Cineform doing much better (around 2x smaller), in this case the file was only ~25% smaller (Filmscan vs. ProRes HQ). Doing a frame subtract comparison with a curve pointing straight up (to maximize error visibility) the Cineform delta was brighter, showing a higher average difference. What is apparent is that while the average difference is higher, there's no macroblock artifacts in the differences as there are with ProRes. Both files are visually lossless before the curve is applied (black screen after the subtract). To fairly compare codecs, especially when they are VBR, requires a lot of tests and PSNR+SSIM to really understand the differences, since the results are visually lossless. What I can say is the nature of the differences between wavelet and DCT show that wavelet differences are more pleasing (as expected), since there's no macroblock artifacts added to the output.

Folks on Reduser discussed this where one user reported at least 30% smaller vs. ProRes: http://www.reduser.net/forum/showthread.php?121568-Cineform-superior-alternative-to-ProRes. In limited testing for high frequency and low frequency input, I've seen Cineform 25% - ~200% smaller vs. ProRes. The nature of the changes after compression are more pleasing: generally a softer image vs. macroblock artifacts which will look like aliasing when at high levels. Both codecs are visually lossless for many round trips.

Please post the PSNR (or SSIM) link if you can find it- curious to see what the statistical differences are.

EDIT: tested a clip again with lots of background blur, sure enough Cineform was significantly smaller: 2.3x. This matches what frame.io reported almost exactly (2.28x).

In summary, for high frequency (high detail) frames, Cineform is around 25-30% more efficient and with lower detail frames can be 2.3x more efficient (probably more as blur goes up and detail goes down, e.g. with motion blur not just background blur). Try it yourself!

 

[Cary Knoop]

What I can say is the nature of the differences between wavelet and DCT show that wavelet differences are more pleasing (as expected), since there's no macroblock artifacts added to the output.

I agree with that.

Personally I like Cineform over ProRes because it is open source, and it is efficient as well. But it does not cure the common cold.

Please post the PSNR (or SSIM) link if you can find it- curious to see what the statistical differences are.

You can do it yourself with FFmpeg, you can compare (the original) with the test one using PSNR and SSIM.
Vapoursynth has a VMAF filter (a filter based on Netflix's perceptual quality algorithm) which might be interesting as well.

For fairness sake, it is important to start with an (almost) uncompressed source, because a compressed encoding will have a slight advantage if the source uses the same compression method.
Since both codecs are all-intra you effectively can use single frame images as examples.

 

[jcs]

I agree with that.

Personally I like Cineform over ProRes because it is open source, and it is efficient as well. But it does not cure the common cold.


You can do it yourself with FFmpeg, you can compare (the original) with the test one using PSNR and SSIM.
Vapoursynth has a VMAF filter (a filter based on Netflix's perceptual quality algorithm) which might be interesting as well.

For fairness sake, it is important to start with an (almost) uncompressed source, because a compressed encoding will have a slight advantage if the source uses the same compression method.
Since both codecs are all-intra you effectively can use single frame images as examples.

Thanks for the friendly debate, Cary, I've done enough tests to be confident with a ~25-200+% efficiency advantage for Cineform over ProRes along with more pleasing compression results (see edit above). I learned that there's a significant difference in performance based on scene detail. I thought you had a link handy for a PSNR example. From my delta-gamma test I'm confident to use and recommend Cineform (and JPEG2000) over ProRes (and DCT), especially knowing that as background and motion blur goes up, Cineform does really well. Something else I learned is that dropping down to Cineform Medium and Low didn't drop file size much for the high detail scene: a nature of their VBR implementation. I don't recommend JPEG2000 (I think it's actually required for DCP) since currently the compressors and decompressors are slower than Cineform. I previously didn't recommend Cineform, even though I bought it years ago for stereo3D work, because it was extremely buggy. Now that it's open source-you can download it from Github, the bugs have been fixed and it works really well.

 

[Martin.G]

Sorry if I'm going to come off as salty, too. That's not my intention either. Just a little on the defense regarding machine and OS choice! I'm just giving you the advice I would give to a best friend. Take it for whatever it is worth, but it is offered in friendship.

No worries. Tho im a little shocked you pay 4500 USD for a machine.

A gaming computer at half the price would do the same job just as good. The only advantage is the Mac OS with certain apps that is not available on the PC. Like said Prores. And for me, im not paying a 2250 USD premium for a single codec.

I might do 100 clips to Stock sites this year, spread over the full year. All of it just nature and other specific things that wont sell very high.

We used Apple comps in film school. I was not impressed.

 

[Publimix]

No worries. Tho im a little shocked you pay 4500 USD for a machine.

A gaming computer at half the price would do the same job just as good. The only advantage is the Mac OS with certain apps that is not available on the PC. Like said Prores. And for me, im not paying a 2250 USD premium for a single codec.

I might do 100 clips to Stock sites this year, spread over the full year. All of it just nature and other specific things that wont sell very high.

We used Apple comps in film school. I was not impressed.

I agree, ffmpeg is free so you can output Prores for nothing. And when you are handy with a pc it can be done automatically too. You can make the software transcode everything you put in a certain folder.

But be aware, don't try to argue with Apple owners, they'll never forget..