AIC-M2t and compression problems

[Eric Bilodeau]

As I shot tests friday night and I had the camera for the weekend, I tried a few things to digitise and edit my rushes. I came to a very problematic situation: imovieHD takes the m2t and converts them to the AIC (Apple Intermediate Codec). This codec is heavier than the original m2t but is clearly very destructive. I edited a first rough cut and tried a few color corrections. The end result had a lot of compression artefacts. Also, as suggested I had a timeline in FCP with upper field priority and the fields seems to have missaligned at some points for a travelling shot. My original file was in CF24. I recaptured the rushes in m2t using DVHSCap and I will prepare a few tests to show you the degradation occuring from the AIC codec. As imovie accepts only AIC clips, I may have no other choice than to output a final edit in AIC to be able to send it to the camera via imovie. Still it will be only one compression. As I progress with my tests, I will update this thread.

Chris, could it be possible to post some clips and stills of the test charts on DVI?

[Ben De Rydt]

Bad news.

Where the compression artifacts visible in the canvas, or in Cinema Preview?

[Eric Bilodeau]

They are visible from the canvas to the final output to camera, even downconverted to SD.

[Steve Connor]

AIC is not very promising at the moment, I have tried very hard to find a workflow that will improve the quality of it and every time you get the same artifacting. I really hope that FCP5 has a different way of handling HDV, or Quicktime 7 vastly improves the quality of the codec.

It's going to be a long two months until NAB!

[Eric Bilodeau]

I have a good example of the awfull compression occuring with AIC multi-degenerations. I would like to post the file here if possible (Chris? is it possible?), it is a DV quicktime file with 4 windows: 1 for the original m2t, one for the "original" AIC captured by imovieHD, a 1st and second generation AIC compression. The images are already in low key with a bit of noise on the original footage.

[Bryan McCullough]

Can't DVHSCap export back to the camera?

[Eric Bilodeau]

I don't have a camera to test it right now but yes, it can with a compatible m2t. MPEG streamclip, by squared5 is able to create multiplexed mpeg streams from quicktime files, probably a compatible one to be able to export back to the camera but somebody else will have to test that.

[Eric Bilodeau]

The compression examples are now posted on the forum at this adress:

http://www.hdvinfo.net/media/ebilodeau/

Description of the files:

AIC_compression.mov
The compression degradation example in DV (for easier viewing on a tv)

still_stairs.ts
The shot used for the AIC compression example, a low light, already artefacted (lightly) shot.

Thanks to Chris and Jeff once again for hosting them :)

[David Newman]

Very interesting. I added brightness, constrast and saturation to clearly see the artifacts AIC is introducing. I haven't had time for my own investigation, but clearly this download shows AIC many flaws.

What I see :
1) 16x16 macro blocks where the chroma disappears in the center of the image (16x16 makes sense for a 4:2:0 DCT compressor.) Although normally DCT does better than this for multigenerations. Did you add anything to each generation?

2) Classic green shift. This is a problem with truncation type rounding in YUV codecs. With each generation, the YUV value may get lower (due to truncation) so the color will shift to green (YUV of 0,0,0 is an illegal deep green -- we call it the green slime.) This could be a failure of color space conversion not compression (depends if you were able to maintain a YUV workflow.) In any case green shifts can be avoided (years ago Microsoft's DV codec had a major green shift.)

3) Interlaced fingering. For speed I'm guessing the field are compressed separately. When different artifacts occur in each field, it shows up as lots on 8 pixel long horizontal lines (not present in the original.)

Note: I have spoken with someone in the Apple codec team about AIC. He commented that consumer AIC testing my result in less than desirable results as Final Cut Pro as it is not yet ready for this workflow. I don't understand the issues, but things may improve for AIC in the future. But from a CineForm prospective, we are not worried.

[Eric Bilodeau]

David,

Very interresting report. Here is what I did for the test:

1-import AIC through imovieHD.
2-import TS file from camera using DVHSCap.
3-convert TS to uncompressed using MPEG streamclip.
4-Imported the AIC file in FCP and recompressed it directly to another AIC.
5-Imported the new AIC to recompress it into a third AIC.
6-opened a DV project in FCP.
7-Put the 3 AIC and the uncompressed onto the timeline, cropped but in a 1:1 pixel ratio (100% size 1440X1080) of the same area from the original file.
8-Exported the DV file using export / QT movie.

So I did nothing else than recompress the file, no color correct or filter used.

[David Newman]

Eric,

Clearly you have done everything correctly. We will have to see what Apple does to address these quality issues.