artefacts in plain colour

[gfunk]

Dear everybody,
first of all - great thanks to the developers for this amazing free DCP-creaton-tool!
Now my question: My material is probably a bit special. It is an animation which contains only plain colours. Only full black, full white and full red. After conversion to dcp in some parts of the film the red colour gets yellow artefacts as you (hopefully) can see on these images
https://www.dropbox.com/s/pb6dj3ksxymum ... P.PNG?dl=0
https://www.dropbox.com/s/3643kx5xjoe98 ... P.PNG?dl=0
I'm not sure whether it might be a player issue, but since both EasyDCP and NeoDCP show similar (not identical) artefacts I do assume, that these errors might be compression artefacts.
I have tried following things:
- Putting a noise all over the image. This only turned the yellow artefacts into yellow noise
- Putting the JPEG2000 rate all up to the limit (250 Mbit/s). This did not remove the errors. Probably reduced them also making the playback less fluent on the same machine.

Thank you in advance for help and feedback!

Gerhard

[Carsten]

Hmm, they are not even easy to see for me. Can you make one of the original images available, before J2C compression? J2C involves Gamma and color space conversion, and sometimes things become visible that have been invisible in the original images, yet are there.

- Carsten

[gfunk]

Thank you for such a quick reply!
Yes, also on my machine the errors are hardly visible on the secondary monitor.
Here is a screenshot from VLC player playing back the source file (.MOV using the codec "Animation")
https://www.dropbox.com/s/05onqr81chevr ... n.PNG?dl=0

[carl]

If I remember right you need to click the "R/G/B circles colour" button at the bottom of the EasyDCP player window to get it to convert from XYZ back to RGB. How does it look if you do that?

[gfunk]

here you can see the difference between the settings with and without conversion XYZ -> RGB
https://www.dropbox.com/s/ayk40j4sbi80l ... s.PNG?dl=0
https://www.dropbox.com/s/21pmbpzf2cev2 ... 2.PNG?dl=0

[Carsten]

How big is the original source file? I would really like to get my hands on it in order to test it (and view on a DCI system). Because, if these artifacts are not there in the source, we would have to assume a serious flaw in the J2C compression. Since DOMs J2C has been optimized recently, there is a potential cause for issues, so, we should look into it.

- Carsten

[gfunk]

The source file is about 1,5 GB, but i guess, you don't need the whole film (7min). Instead I've rendered a short representative passage (about 120MB) containing the frame shown in the previous posts:
https://www.dropbox.com/s/5c3mngd4cdfkw ... n.mov?dl=0
Thank you for giving me a note after downloading, so i can take it offline again. I'ts kind of sensitive content.

And thank you guys for your support!

[Carsten]

I just pulled it, maybe Carl will have a go, too. You can take out the drop-box link from the above message and send it to Carl via email as well.

I used to do some things with the QT animation codec, and, if I remember right, it is not as clean as you would think. But let's see...


BTW - nice flic, did you do that yourself? How? Ah, found your site. I like it!


- Carsten

[gfunk]

Thanks for the compliments Carsten!
Yes, the animation is my work. The sequence with the rays was originally coded in processing
here's how it looked then, nothing special basically, just sine waves animated with some easing-in/-out
http://gfs-gifs.tumblr.com/post/103915453596/a-v
afterwards - distorsion on it to get rid of the super-clean looks, that's it.
The rest is drawn frame-by-frame.
You can have a look at the whole 7 min, if you like. Just tell me where to send the vimeo link+password.

I was expecting some kind of obstacles in the comression with this radical black-white-red look. And actually I'm not bound to the Animation Codec. The whole material exists as a clean losless TIFF-Sequence, but I just haven't found the right workflow to put that material into DCP-o-matic

[Carsten]

I just started some tests with this, and so far can confirm that your source footage is as clean as it can be - all full white, full red, full black, except for the antialiased edges. I can also confirm I see quite some compression artifacts, ringing, etc. when I open your screen-shots from EasyDCP and NeoDCP Player. Now it is possible that software players use shortcuts in decoding, different wavelet quality levels, etc. I'll check various encodings of your source footage myself later in the cinema, and I will also unwrap some stills and have a look at them.

But so far it seems there is no noticeable difference in encoding between old 1.83 and new 2.71, so, I don't think the J2C optimizations came with a real penalty.

- Carsten