v1.0 (kinda anniversary) release discussion

[ilia3101]

We can be a maybe a couple days late with this release??

This will be place to discuss this release for now.

Maybe we can come up with some other ideas for it...

[bouncyball-git]

Skip beta stage alltogether? ;)

[masc4ii]

Until now I only did some bugfixes. Yesterday I started in analysing and trying out making dual iso faster. But until now I had not to much success: I multithreaded amaze interpolation and tried to comment out other parts of the algorithm to see if something gets faster... in a 3min export it was not a single second faster 😢

[masc4ii]

Not to difficult should be to add the gradient filter. But render speed would be a bit less than half: we process the picture twice (1x original, 1x with gradient settings). Then a alpha mask should be calculated and the pictures must be blended together. But I am not sure yet where to add which parameter structure into Ilias code to have it nice... (and maybe also in a way n gradient filters could be added in future). This feature should be discussed here https://github.com/ilia3101/MLV-App/issues/49

[masc4ii]

If we could add BMD Film profile this could also be great, because of some EOS Luts. But I haven't found any function therefor yet. Even trying out with the Resolve LUT linear->BMD looks not correct using our linear profile (looks very different to what Resolve shows).

[bakersdozen]

I see MLVProducer has BMDFilm curve, could be a starting point or reference? This could help eliminate Resolve from some people's workflow if they wanted to. Shoot > MlvApp > NLE

[masc4ii]

I see MLVProducer has BMDFilm curve, could be a starting point or reference?

Two problems: MLVProducer is not open source and also seems not to be correct here (for me this looks very different (both BMD color space + EOSHD LUT)):

This could help eliminate Resolve from some people's workflow if they wanted to. Shoot > MlvApp > NLE

Hehe... this is what I do nearly all the time 😏

[bakersdozen]

Good point, thanks for doing the comparison too, I thought maybe it was a not 100% version of the curve. It would just be awesome to be able to not rely on resolve and have MlvApp all the way! You get it.

[ilia3101]

Actually, there seems to be much of the MLVProducer's processing source code on sourceforge where it's hosted, there is both c and basic files, though some of it looks obfuscated (weird variable names with $ in them).

[masc4ii]

Really? I was searching and always found executables and no code. Could you provide a link?

[ilia3101]

I only noticed this yesterday, I looked as I'd seen reddeercity say it is open source. Anyway I have no clue about the license so it might not be much use and it's really confusing to read

[ilia3101]

Ok I'll send a link 1 sec

[masc4ii]

Found it: https://sourceforge.net/projects/mlvproducer/files/src/ But all curves are in a binary format... :(

[ilia3101]

Maybe there could be something useful if you look

[masc4ii]

All the magic happens in the basic files. But there are no comments - so very hard to understand. BMD color space is realized using a .cslut ... a kind of LUT, but it seems it is a format by awpstar. I have no idea what he is doing there.

[bakersdozen]

Don't know if these will help. Unable install the Mac file and don't have access to PC right now. BMDFilm for ACR.zip

[masc4ii]

Thanks, but this seems also to be a kind of binary format. What we need in the end is an equation like in the end of this post: http://www.magiclantern.fm/forum/index.php?topic=15801.msg158145#msg158145

[bouncyball-git]

I don't think we get equation ever. We could try to mathematically approximate to the curve but we even do not have reliable real curve ;)

[masc4ii]

Right, I think so too. Resolve comes with a LUT Linear->BMD, but applying this one don't looks right as well - so why doing the approximation here?!

[bouncyball-git]

Here are 2 curved BMD Film and BMD Film 4K, but who knows are they right ones or not :)

[bakersdozen]

I decomplied the VisionColor BMDFilm for ACR .dcp from the files above for Canon. Using dcpTool. Here is the xml output, i'm not sure if this will help or not. See the section.

xml.txt

[masc4ii]

Looks like a tone curve... thanks. But the variables are different from what I've seen until now. Let's see if that helps ;)

[masc4ii]

Goooood... I added 3 new debayers to the app and made some bugfixes... I would please you to test it a bit - when you have no findings and no other little projects for MLVApp this could be v1.0 ?! 😄 And could someone film a cake with a "1" candle for a new screenshot?! 😆 I could only fake a picture...

[masc4ii]

[bouncyball-git]

Great photo we can "photoshop" it to the MLV App main screen :) I've tested debayer methods and they are working great. Simple is a bit faster than bilinear LMMSE is 2x slower than threaded Amaze

Some quirks with OpenOMP optimized dual iso (to be debugged). Sorry about not being helpful. Your effort is huge!

[ilia3101]

OMG really amazing stuff going on!!! I now have a lot more free time as it is summer. What I've been trying to do for this anniversary release is basically the white-balance branch, but to actually succeed this time. Working on it right now. But this is truly amazing what's going on here.

I've gone through a few ways of doing it - the most difficult thing is keeping highlight reconstruction working while avoiding clipping everything. If the processing was floating point, these problems wouldn't exist :laughing:, but that not worth doing considering how slow it is already. @masc I remember I mentioned trying out integer multiplication + shift for matrices - I did it and didn't notice a difference though I may try again, and use it in more places in processing. I didn't do any actual timing tests.

Also apologies for all the difficulty you guys may be having maintaining the code, I'm having a hard time too. There's some things about it that I did a year ago that would be good to change, but there is not enough time in the universe to do that.

[ilia3101]

I could try and make a cake in blender, not promising it will be realistic but it would be amusing.

[masc4ii]

Hej Ilia! Nice to hear you are back to MLVApp! Sounds cool! I also tried out to calculate the processing in another way... rendering speed was identical. So I also think we can let it as it is - or someone has a better and faster idea 😄 Don't worry! Your code is good! I think I mostly understood what you did now. That is also why I tried out a lot new stuff with it. I hope it is no problem for you...

A blender cake sounds also well, haha.

[masc4ii]

BMDFilm profile is there... yeah... I realized it different to the other profiles. But it seems to work and looks correct in my test (please try it out too!): MLVProducer: Davinci Resolve:

MLVApp:

All three with only set WB, BMDFilm profile and EOSHD LUT.

[bouncyball-git]

Fantastic man! MLV App image is a bit cooler (on the bluish side) I guess it is WB.

[bouncyball-git]

Is "bmd_film.h" linear to bmdfil lut? is it 4K lut or older one?

[bakersdozen]

Awesomeness MASC. Will compile when I can and test bmdfilm. Great work, would also like to know if 4k version or not...

[masc4ii]

Japp, bmd_film.h is the converted linear-to-BMDFilm.cube :-D (I know I won't get the nobel price therefor, but hey... working). It is not the 4K version... there color is again different, so I thought the non 4K should be better for most non 4K footage (whyever color must depend on resolution, maybe you could explain).

[bouncyball-git]

It designed for sensors used in Black Magic cameras. For older cameras with <4k output and for newer cameras with 4K+ output.

I think non 4K BMD Film lut is more appropriate for canon sensors, however I can be wrong :)

[masc4ii]

Description (under releases) and picture is ready... now only the last bugs have to be fixed 😄

[bouncyball-git]

Love waveform :smiley:

[ilia3101]

can we put this in the pic for release to be funny (not necessary on readme tho unless some of you think its a good idea)

[bouncyball-git]

Bayer pattern cream is nice ;) I like both, so you decide.

[masc4ii]

Haha, very good Ilia! Very cool! I also like both. You're still the boss... so I also think you decide 😄

[ilia3101]

Okay let's use this everywhere until next replacement comes along

[masc4ii]

I love it!

[ilia3101]

wow cool label

[masc4ii]

What do you think about releasing v1.0 this weekend? I don't think we solve the windows malloc problems in the next days. For dual iso I would revert the openmp lines again, but after the release we put it in again and try to debug it in the future. What do you think?!

[ilia3101]

Sorry I have been absent, currently on holiday. Sorry didnt manage to contribute much to 1.0 (Doing another project too, potentially helpful to MLV app, though I don't want to talk about it too much because when people tell me they like an idea I lose motivation to actually do it). Was going to say let's release it but just noticed it's released, yay!

[masc4ii]

Sounds great... hrrrrhrrrmmm... not great at all... (hope that helps your motivation). 😆