Messages

Ok I did get 5D3 with broken SD slot. ML works great. Gonna replace it myself, the part is 25$ on aliexpress.

I have downloaded the newest crop_rec_4k build, which is from july 2018, and all the presets are great fun to test out, however I have noticed the UHD 1:1 mode is not centered vertically:



With swirly bokeh lenses this is quite noticable

Can anything be done about this? I tried the advanced settings in crop_rec, but only succeeded in making the image purple and glitched. I am ready to investigate, make changes in code and recompile if needed.

Hey, I saw the github issue about a new logo. How about a homage to the horse in motion? There's some nice icons here, as inspiration.
I don't like much the idea of using magick references... it was overused by many already (imagemagick, g'mic, etc), IMO.

My thoughts:

- You have a good point about the 'magic' thing

- I want the colour scheme to be based on mostly Red, Green and Blue (because bayer filter has those colours)

- I don't want it to look like an iPhone app (MLV App's name containing 'App' is not enough to justify that)

- I like reflections and 3D-ness, I also the icons in macOS

I would like a new icon.

My original logo is not amazing. It does not relate to MLV or video or image processing in any way, but I think it looks quite nice. And because it is made in Blender you can do stuff like this:



(from 1.0 release)

Yes its a joke, and you just made it seem so much funnier

Noob 5D3 questions coming up ...

How important is having a working SD card slot for the 5D mark III in 3K/4K?

Does having magic lantern on the SD increase performance of the CF? Can you even install ML on the CF instead of SD?

Also does the feature where it uses both cards at once for increased write speed still exist?

@bouncyball glad you are going to investigate this. Can't wait to find out what causes that error so I can also know to avoid it in libMLV.

A raw-based "RGB" readout could be useful as well; the current one is YUV-based. Once that is done, adding XYZ / xyY to the spotmeter would be probably straightforward.

This would be amazing!

Curious how Davinci resolve implements optical flow. Do you know @masc?

As masc said, on "debayered picture data"

But I have suggested a while ago, we can try to interpret bayer as 4 channel image, which would allow optical flow on bayer raw. Haven't tried it of course, but that's what they do for neural networks.


Also who noticed the new url?

Cameras are not very good colorimeters (or whatever u measure colour with), but mostly close enough for non science purposes.

Kelvin and G/M readout near the spot meter square.

This isn't a very meaningful reading of colour. Do you just want that, or would you like something more meaningful? Like XYZ/RGB or xyY values? I can help with that.

Most severe case of Adobe look I have ever seen

Just a thought.

Don't forget some users may be shooting with an IR converted camera, so colour correction, eg WB, will need extreme shifts and channel swapping.

As a say, just a thought, if one is trying to make this App a killer App 

True, maybe we could have an option where one can enter a custom matrix, which can do both channel swapping and any kind of extreme WB.

Or maybe a raw channel gain option, would be simpler.

I'm not against adding colour wheels, I tried to do it before. But definitely never liked any effects they create.

Can you make a color grading not changing colors, saturation, etc.? If you change anything it's already "inaccurate".

There's many things that you can change and it's still accurate, I even think you must change stuff to increase accuracy, as monitors have very little dynamic range compared to real life, and you are trying to recreate the way it really looks, but within that very limited range of brightness. I think curves/contrast are necessary, and a little saturation can be as well.

Another example - white balance. Temperatures other than 6504K on the slider makes colours not be their true values... but it only increases accuracy. The intention of white balance is to match the scene's "white" to the "white" of your surroundings (meaning your monitor's white), which is the white colour your eyes are adapted to. It lets you see the differences between colours correctly.

Found the answer in mlv_blender, xyz to rgb.

Every camera/mlv has an XYZ to (camera)RGB matrix which we must use to convert camera native RGB to any other colour space. Notice it's XYZ to RGB, so it must annoyingly be inverted first, then we can convert from XYZ to some standard RGB space like sRGB or rec2020... You can choose this in processing profile settings.

Also a L*a*b stage could be added for any processing effects that need it.

I don't think there's any hope for accuracy once you start using colour wheels

I have seen many of these (memories)

The EOS M 3.3K/3.5K ones - are they shot in anamorphic or can you actually get that much resolution on eos M?

Might be the video player you are using. Literally 99% of them fuck something up in terms of displaying colour, including apple Quicktime.

Or maybe you have calibrated your screen, and the video player respects that (unlikely), but MLV App does not (this is an issue on Windows and linux)

edit: also try some different formats

Right. The issue is that most people just use their iframe, instead of the direct binary link, which requires javascript.

Ah ok, good to know.

edit:Possible change for 2.0 release?

Hopefully soon actually, it's not too big of a change.

@Dmytro_ua Thanks for the comparison. Looks like ACR handles that pizza sign so much better. This kind of stuff I really want to fix for 2.0, or possibly MLV App 2.

In your MLV App edit, I think if you enabled highlight reconstruction, reduced the Light strength slider closer to zero and pulled the highlights down a little more, the background might look smoother (less bright). What tonemapping did you use BTW?

Missed all this

Well, not for me and a1ex:
https://www.magiclantern.fm/forum/index.php?topic=18003.msg216880#msg216880

I know IBB is not great, but I have simply been embedding the image. This way I assume there is no javascript being run, right?

If I was to judge:
- Your version, @Dmytro_ua, is the most natural posted until now. I would just lift the shadows a bit, to bring more details to his overcoat.
- My version was purposefully exaggerated in the teal/orange. Noise everywhere. Bad separation of midtones in his face. Too much saturation.
- @Bender@arsch version has muddy midtones. I would bring more exposure there and decrese exposure in deep shadows. Colors are good.
- @DeafEyeJedi has some issue with magenta/green. Skin tones look pale.
- @Ilia3101 has a overall green cast.

Is your's edited in Rawtheapee? The denoising look is not something I've seen before from mlv app.

And I mostly agree with your judging. Mine should be less green, but all of everyone else's is a bit too magenta. Who used what software? I notice mine preserved the chromatic abberation (green edges) of the bokeh more than the other edits, which I like.

DeafEyeJedi's looks that way because of using rec2020, the shape of the gamut is not greatly shaped for grading like it's rec709 (green is way out there), AdobeRGB and Alexa RGB look better in my opinion (this is why almost every type of grading workflow can get good results with Alexa footage).

Also I just remembered, MLV App tint adjustment is quite crap actually, it is not done in the LMS colour gamut with white balance, but right after, when the image is in the final gamut. I did it this way as it easily preserved old tint code. I thought this would not make much difference as tint adjustments are generally small, but seems like these days everyone is adjusting the tint a lot.

Glad to see that the good old 5D3 is creating so much inspiration, guys.  Thanks a lot for your feedback.

Ilia3101,

If you consider buying the 5D3, now is the time to do that.  Thanks to the mirrorless madness, prices of this fantastic camera have dropped dramatically.  In my country, serious photographers are so pleased with their 5D3-s that they use them until the shutter breaks - very often after 350'000 actuations, and then they just throw them away since repair is too expensive.  Then they look for another used 5D3.  If you can find one in a good condition and with a low shutter count, do not hesitate to buy it.  You will be very pleased with its performance and especially the unmatched video capabilities and quality with the latest experimental ML builds.  Also, it will provide excellent service to you for many years to come.  So, good luck to you and everyone else considering a 5D3 purchase.

Any of those photographers can throw their 5D away to me if they want I think I might buy one with a high shutter count as it's really cheap that way (high risk, high reward lol), and I don't really care what the outside condition is.

(Sony raw will have to wait).

What does this mean

As far as i understand they reconstructing a 3D scene from the image and applying color corrections dependent on distance?
And " In all cases, the simple contrast stretch S1, which is global, works well when scene distances are more or less uniform".
So yes, this is something more, but can't much inspire someone more or less familiar with the subject.

IDK about the youtube vid, but in the paper if you look at it, they are using the depth map to undo the spectral absorption effect of water, also taking the camera's spectral response in to account. It reverts the depth colour differences within the scene quite accurately.

It's very smart colour corrections. Will it look better? Maybe... maybe it won't even make a difference.

Saw that, very genius and inspiring.

Was it shot in crop mode? maybe the crop window was in a different place, not sure how to help with that. Please uplpoad a file if possible.

YCbCr according to this:
http://lclevy.free.fr/cr2/#sraw
edit: it is encoded in YCbCr 4:2:2 and needs interpolation for the Cb and Cr values.

Thanks

I'm reading this right now:
https://github.com/lclevy/libcraw2

"Libcraw2 has two kind of interfaces:
an C API, mainly used by crawtool, to offer CR2 data handling in a Camera agnostic way
an Python extension, PyCraw2 a simpler API which goal is to offer all required functions to render RAW data into a 16bits RGB image, either from RAW (RGGB) or sraw/mraw (YCbCr) data."

Looks liek a nice library. Still need to decide on an official way of including other types of image data like RGB in MLVs, only bayer images are properly supported.

I still haven't ever done dark frame, curious to see how well it works.

Does anyone know what kind of image sraw and mraw is? I'm quite certain it's not bayer. Is it RGB or YUV? (I think i've heard something like that)

@cmh Interesting. Looks like they have the correct colour matrix for the mark II in that JSON... so it's a mystery why rawtherapee gives such wrong colours

Yeah just use the lens cap, and be in the dark if possible.

Could take a few pictures and average them too.

Is this in video or photo?

If it's video then take a short dark video instead of a still.

Looks like fixed pattern noise to me, plus significant underexposure. Try using a darkframe (create one with the same settings, shutterspeed and iso)