Messages

@berechiah: Here is explained what to do: https://www.magiclantern.fm/forum/index.php?topic=16054.msg215358#msg215358

Have you read and understood this post?

This is the sensor:
RGRGRGRGR...
GBGBGBGBG...
RGRGRGRGR...
....

Now read each column (=1x1=no stretching needed):
RGRGRGRGR...
GBGBGBGBG...
RGRGRGRGR...

Now read each 3rd column (=1x3=needs 3x horizontally stretching in post)
RGR...
GBG...
RGR...

Now read you suggestion, each 2nd column (=1x2=needs 2x horizontally stretching in post)
RRRR...
GGGG...
RRRR...
But where is the blue gone?! ---> impossible.

On the other side, 1664x2126 is already to large to get continuous recording. How do you want to save the double data at the same time?

1) Both camera's mustn't be cropped to record full sensor @3x3 + audio. 5D2 has a slightly higher resolution in this mode and produces a little less aliasing.

2) Best use the link from dfort. @12bit "near to FHD" 5min on 5D2 are about 21GB+some MB for audio. On EOS-M it can be less when using lossless compression.

but it is already stretched to 16:9. it just uses a lower resolution. Can we not stretch it to 16:9 but using the resolution in the 2.35 setting?

There is nothing stretched. It is cropped!

for the anamorphic mode, would it be possible to have a stretch of .5 instead of the .33 so it doesn't upsample as much.

Physically impossible. You can read each column or each third column. Otherwise you'll never read in some colors. Look at a bayer matrix... This is also why there is a 1x1 and a 3x3 mode, but no 2x2.

If you are able to compile... it is commited. Please test and report results. It is more complex than it looks like - no idea what else could go wrong

Like that?

Thanks Ilia.

For getting something very close to Fattal Tonemapping, do the following:
- Highlights: -100
- Shadows: 100
- Contrast: -100
- Clarity: 100

That's right, MLVProducer does a perfect job in removing pink highlights. Seems to be a different method. Unfortunately I haven't understood how it is done - nearly no comments in code

Do you think it would be possible load profiles automatically using lensfun (based on metadata)? Also, this lensfun software seems really cool, might be of some help in MLVApp.

Yes, lensfun is really nice - but very huge. And I did not found out, how to use it. But what I found out: have a look here, you can see that it won't correct CA's for most lenses. And I am not sure if it would when using a Speedbooster on a EOS M (which produces a lot of CA's in another way than the lens would without).

The "DR Compression" in RT is actually a tonemapping operator. Specifically the Fattal method:
http://www.cs.huji.ac.il/~danix/hdr/hdrc.pdf
It is a bit different than what MLVApp does in profiles. The way I see it is: where the profile is used to "comply" with a standard (e.g, Log-C) in a linear way, the tonemapping tries to compress or lift the illumination dynamically.
I'll see if I can test the vid.stab and the CA next week. Also would like to test the ACES branch

Fattal is in theory not far away from Reinhard Tonemapping. And this you can choose in Profiles comboxbox. In ACES branch we separated gamuts and tonemapping. But in current master branch it is still mixed.

@andy kh: this is a known issue with all ML cameras. It happens not always, but if it happens you just should lower RAW white level. We also wrote this in the manual https://github.com/ilia3101/MLV-App/wiki#highlight-reconstruction

Maybe white level should be set according to this clipping point instead in cam.

I already tried this, but without success. No idea what happens with less than 14bits in camera. The identified value was mostly way to high and the pink was still there. Looked as if the clipping point was different for each pixel and each frame

RawTherapee... you never can say for sure if any of those modules are compaible with our software. In general it isn't at all out of the box. The first time I found some CA correction code from RT was in february, now I got the code adapted to be able to compile and use it. I was very happy got this to work, without knowing how it works and what it does. There is no documentation (or I haven't found it).
In the current official MLVApp version DR compression is done behind profile combobox. I don't expect it even to be a little similar.

For all guys who know how to compile MLVApp:
I had success in adding the chromatic abberation correction from RawTherapee into MLVApp. This means, an older version of it. The older version has a code interface very close to what we need, while the current version is so compilacted, that I had no success in just compiling. What you get are two new sliders to correct red and blue CA's. Moving the red one to around 0.7-1.0 fixed most CA's in my test clips. Especially for all EOS M guys with Viltrox Speedbooster, this is a great addition, because the booster produces so many CA's. Here an example - not perfect, but better than without.




Would be nice if someone else could test this...

Same for our vid.stab problem. Would be nice if someone could test and report what works and what doesn't.

Thanks for your help!

Since february I don't get any notification from ML forum. Before it was working just fine. My spam folder is empty. Any idea's? Someone else has this problem?

Check wav(audio) routines. When enabling Export Audio files comes out. Without it files are blank(checked prore422 and prores444).

Thanks Danne. But exported now without audio... empty file, ffmpeg crashed and pulled MLVApp down. Somehow vid.stab doesn't like RAW to ProRes...

Don´t be sorry. Interesting bug. Any idea what it might be? How is the vidstab setting applied? I only ran the script with mov files in Switch but always working:

Code Select Expand

ffmpeg -i input.mov -vf vidstabdetect=shakiness=10:accuracy=10:result=transform_vectors.trf -f null -
ffmpeg -i input.mov -c:a copy -c:v prores_ks -profile:v 2 -pix_fmt yuv422p10 -color_primaries bt709 -color_trc bt709 -colorspace bt709 -vf vidstabtransform=input=transform_vectors.trf:smoothing=10,unsharp=5:5:0.8:3:3:0.4 output.mov

Could you point me to code? Maybe the the trf file is erased too early or something? Only guessing...

@Danne: have you tested this with a ProRes mov or a RGB-avi file? Did that work? I think, even if MLVApp does export a ProRes with vid.stab enabled, there is no staibilizer applied, while exporting h264 it works fine. I remember I had similar issus in analysis pass: it only works with h264, so I implemented the pipe to pipe thing. Maybe here we have the same problem, but pipe to pipe means quality loss (not good).

@Bouncyball: Picker is no problem... functionality is there, we just need another icon -> easy exercise. But we need some buttons for adding/deleting pixels, for loading/saving a pixelmap. We should think about another widget. 

The more I test, the less I can tell you what's the problem. Now it feels each file may crash or not, with same settings, 100% random. That really sucks.
Exporting 5x exactly the same will lead min. 1x to crash, if vid.stab is enabled. Here a 5D2 file.
https://www.dropbox.com/s/b8wo39wwdbboqkb/M05-0608.MLV?dl=0

Still don't understand the problem.
--> Would be cool, if some mac users could try it out and post what configuration (Camera, Resolution, Export codec...) combined with vid.stab enabled works or not. 99% of my clips work, e.g. EOS M MCM rewire at 16:9 creates empty files for ProRes. ffmpeg stopps working when starting final output. All other files I tried here work just fine. 
Windows users must compile latest commit to be able to test. But on windows ffmpeg seems to play even more Diva.

vid stab dont work at all after export ready i could not find any video in the folder if i check enable vid stab

You are on Windows? I commited a fix today for Windows versions... don't know why this single character was missing, because I tested on Windows successfully before. And because of this single missing character in ffmpeg command Windows users get... nothing. Sry.

Thanks for your commands. Will try it the weekend again. I like bugs most, which occur only sometimes...

Hm... some more testing... that happens only for my EOS M clips - 5D2 is fine with ProRes422 output + vid.stab. 

Edit: what the hell is doing ffmpeg?! Sometimes it works, sometimes it crashes. Same command, same input, same parameters. 
@Danne: do you have the same problems with your script stabilizer version?

I am very sorry to say that it seems I have found a bug with the new vid.stab stabilizer: I might have tested to much with h264 (here it works fine), but when I export e.g. ProRes422 or 4444 I get empty files - no idea why yet. Strange things happen with ffmpeg, I don't understand it yet.

 

No ACES

Right, sorry. There is a development branch in the repos, if you like you can try it out. It is named "Better Processing". This one has ACES processing. Unfortunatelly we still have some problems with it - sometimes the output looks way better, but sometimes it looks very strange. That's why it is not merged to main branch yet.

Yes, v1.7 is out now. Main changes:
- Added a simple grain generator
- Improved algorithm and quality for Shadows / Highlights / Clarity
- Added High/Medium option to H.264 & H.265 export
- Added ffmpeg vid.stab video stabilizer for ffmpeg export
- Added bicubic spline resizing algorithm to ffmpeg export
- Drop focus pixel map files into app to install (except Linux AppImage)
- IGV debayer is now multithreaded

See first post in this thread.