Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - cmh

Pages: [1] 2 3 ... 6
1
Daaamn, that's some professional work. The camera did the job fine but you guys put in the work and it shows. I hope you had a good time with the workflow, was the roundtrip because the client wanted h265 at delivery or just for the smaller filesize? You did the grading too? The song is 🔥 btw.

2
I'll upgrade the firmware and test this out.
edit: so far so good, I'll keep it for at least a couple of weeks and report back if there's any problem.
edit 2: the preview seems stuck at real time, I can't get the framing mode (I checked the menu, tested various profiles, uninstalling/reinstalling).

3
Academic Corner / Re: Insane video super-resolution research
« on: May 16, 2020, 08:23:53 PM »
On Windows it is extremely slown people are talking of 5 times slower (maybe this can be fixed idk) and without cuda I can't even fathom how slow it would be.
If I'd make a rough estimation based on their github and various reddit posts, it would probably take something like 6 hours per seconds of videos with my gtx 1060 6gb for getting a really good quality upscaling (10 iterations or so), maybe more.
It's unfortunate I don't have another box hanging around.

edit: I love those AI upscaling posts btw.

4
General Chat / Re: Raspberry Pi High Quality Camera
« on: May 16, 2020, 01:58:46 PM »
https://leopardimaging.com/product/csi-2-mipi-modules-i-pex/csi-2-mipi-modules/li-imx477-mipi-cs/
It's 4 times the price and this is pretty much the same as the hq camera. It works on a old Jetson nano tx1 https://elinux.org/Jetson/Cameras

The driver is not open source and you'd probably need their API unfortunately (but CUDA tho plus it has 2 interfaces and you can hook up 6 with an adapter).

So maybe it would works with ease with the 50 bucks hq camera.

edit: leopard imaging's camera works with a tx2 too.
People are working on upstreaming the Broadcom Unicam CSI-2 and ISP peripheral drivers to support libcamera, that will include the hq camera.



5
Academic Corner / Re: Insane video super-resolution research
« on: May 16, 2020, 10:03:43 AM »
Here's their github: https://github.com/open-mmlab/mmsr
I tried to set up an environment real quick yesterday but I should have used debian stable instead of fedora 32 (it was all borked, I needed to downgrade to gcc 8 and it is a pain).
I stumbled across this 5 months old post on reddit
https://www.reddit.com/r/GameUpscale/comments/dz8j8n/mmsr_esrgan_training_and_setup_guide/

Beware: "Take a week or two vacation. Maybe learn a new hobby or two, get reacquainted with old friends and so something meaningful. Training the model takes a LONG time with a single GPU."

Edit: Centos is probably better suited.

6
Share Your Videos / Re: Short film post Cold Pizza Canon 5D MK2 Raw
« on: May 15, 2020, 05:17:12 PM »
This is a great short, I love it. A true WTF moment.

7
General Chat / Re: Raspberry Pi High Quality Camera
« on: May 15, 2020, 05:02:23 PM »
jestson nano cam faq

This is from e-consystems and concerns their camera (e-CAM55_CUMI0521_MOD an AR0521 sensor).
The Jetson nano works with the v2 IMX219 sensor out of the box.

8
General Chat / Re: Raspberry Pi High Quality Camera
« on: May 14, 2020, 10:48:19 PM »
If if it's the Raspberry Pi VideoCore IV gpu that is hindering the capabilities of the Sony IMX477 maybe another board will do the job.
The Nvidia Jetson Nano paired with the Camera Module v2 (IMX219-200) can achieve 3840x1848 at 28 fps if I'm not mistaken.
I would stick a Kern Paillard 36mm f/2.8 D mount lens on that thing.
I haven't done much research tbh.



9
Share Your Videos / Knock-em-over | EOS M
« on: May 14, 2020, 09:59:01 PM »
https://youtu.be
Canon EOS M with Magic Lantern. Canon EF-M 18-55 mm F/3.5-5.6 IS STM.
5k ana. profile, 1280 x 1636,‬ 2.35:1, 12 bits, cdng upscaled in Resolve.
It's just a grade test. I tried to keep everything natural.
Rec709/sRGB in the raw tab with sharpness pushed to 70 (because youtube) and most of the grade is in between rec709 to Arri, Arri to rec709 CST nodes at 0.65 gain mostly for the luma tonemapping.
In parallel to that a Rec709 to Cineon CST, a correction and finally 35 mm reversal BW grain (that is totally unnoticeable after upload) right before a Kodak 2383 x64 lut by Melara from GMIC's website used at 0.35 gain.
Slight halation, done the lazy way with the glow ofx (colorized to red and the spread setting at 0.02).

10
I def get the flickering on both cdng and prores 422 HQ.

11
Raw Video Postprocessing / Re: Powergrade for Resolve
« on: May 02, 2020, 02:21:43 PM »
Sorry, I've been busy, I love the effect, really reminiscent of film. I find your powergrade less in your face than many or the tutorial available on the internet. Really cool.

12
Haha. I'm sorry downloading after effects trial version requires me to enter my credit card number, nope.

The issue? We probably already found it partly because of you. There's a +0.25 exposure difference between ACR and MLVApp because the exif tag sn't there anymore but it's barely noticeable if you use tonemaping and the other +1 exposure that has to be added is a Resolve thing  (maybe only a Resolve on windows thing, maybe just a windows thing but none of you devs have Windows installed, which makes bug reports a bit complicated tbh). Maybe I need to dig in Blackmagic's forum a little harder.

I think all of us went in great length and have been charitable to each others. I'm totally fine with leaving this topic unless someone needs more explainations or someone else come up this same sort of issue (shoot me a PM if needed so I get mailed).

Edit: but comparing with after effects or premiere is a great idea. Baselight would probably be even better since it's the Hollywood colorists heavyweight software. Nuke, Avid (I might check on those later if a trial is available).

Edit again; did I ever told you guys that Windows is free of charge, the only downside if you don't register it is that you can't customize your desktop.

13
Sure, what are you proposing ?

14
General Chat / Re: Raspberry Pi High Quality Camera
« on: May 01, 2020, 09:52:14 AM »
Yeah I saw that this morning, this is super interesting

15
Raw Video Postprocessing / Re: Powergrade for Resolve
« on: May 01, 2020, 09:50:33 AM »
Thanks I'll check that later in the day.

16
Danne there's nothing to be upset about. If people would have find the same results (which is not the case because it seems really hard to reproduce.. I guess) and said well nice finding but we won't do anything about it... Then a-fucking-men, it is absolutely a reasonable answer. Nobody wants MLVApp to change, me included. At least we would have not lost our time on this. Then if my post have been perceived as rude or condescending it was not my intention. It's just a bug report that has to be confirmed type of thing.
Edit: I'll stick around just in case people want some clarifications. PM is also available btw.

17
I feel like if people would really put an effort into what I suggested in the first place, we might understand each others. MlVApp's prores footage with only the rec709 profile applied vs cdng (not an extracted frame) in resolve with rec709 color space/gamma in the raw tab. Every other comparisons are muddying the water. I didn't made any other claim, ACR was only useful to understand the dng tag. If we can all agree on this part then we could move on.

Edit: yes I understand Adobe created the dng format from Tiff, yadi yada, but it's a spec (like the headers, the legal exif tags etc) not the rendition. As controversial as it sounds, It really is a weird metric when it comes to how it should look in MLVApp (well with the inclusion of raw2mlv it makes more sense, I'll reckon). I'm eager to see what would be the result in Premiere.

18
@danne I don't want to be a contrarian but if you don't change anything appart from using the rec709 profile on MLVApp and export to prores as is. You'll get an accurate rec709 conform footage to compare to (not esthetically pleasing sure, that's besides the point).
Edit : on phone: expect some typos.

Same goes for the srbg profile. super accurate, perfectly conform. Ilia should confirm my claims ofc.

Edit: just saying, the pretone curve thing in Resolve is there for historical reasons, that was used in old resolve by default but it will clip your footage like any pretone curve. There's also other mapping ofx in resolve, like in color space transform for example.

Fun fact (maybe) You can also download filmic aka tangent, Reinhard or Corona's tonemaping functions for Resolve's Fusion tab/og Fusion, but you better transform in linear before, same algorithm as MlVApp but tweakable.I should stop there, since it's irrelevant.

19
Unfortunately I don't have any other monitor to test and it's a cheap 6 bit + frc ( probably like all of us who did those tests). There's no color profile used on windows.


@danne
For the tonemapping settings as I stated, I tested everything in rec709 for consistency.
Rec709 profile on MlVApp without creative adjustment, so no tonemapping.
Rec 709 in resolved raw tab with no Luma tonemapping of a CST, precurve tone disabled.
 
But yeah, apparently ACR is doing things under the hood.

Edit: my bad reddeercity you were talking about the gamma curve of the OS, yeah worth investigating. Sorry I literally just woke up.

20
Also keep in mind that the bmdfilm profile from MLVApp is totally fine, you can keep it but this is not what people do, they export to cdng and use bmdfilm in the raw tab which gives horrible colors and worst than that, they sometimes use Gen 3 or Gen 4 to rec709 as a final CST node and start grading (and this is the reason why I don't want to talk about this and made my comparisons only with rec709, without tonemapping in Resolve or MLVApp, it's complicated enough).
Let me take a screenshot real quick.
Left is cdng with bmdfilm in the raw tab and a CST to bmdfilm Gen1 to rec709. Right is prores with MLVApp's bmdfilm and the same CST to bmdfilm Gen1 to rec709

This is ungraded, just the +1.25 exposure thing and a +15 tint for the cdng clips (this is common even across other colorspace/gamma settings given the appropriate CST ofc, in a sense that if I take the same screenshot with cdng with rec709 in the raw tab +1.25 exposure +15 tint vs prores with the MLVApp's rec709 profile or let's say cdng with linear in the raw tab +1.25 exposure +15 tint with a CST node to rec709 vs prores with the MLVApp's linear profile with a CST node to rec709 in Resolve, you wouldn't be able to tell where's the line, is that clear, I'm not sure).

MLVApp's bmdfilm profile is not a problem, if I compare a rec709 footage from mlvapp (no tonemapping just the plain rec709 profile) and compare it to a prores with the bmdfilm profile and apply a CST from Gen 1 to rec 709 in Resolve, it's close-ish (a little more vibrant and a slight gamma difference but that's to be expected, I guess you got the matrices made for another NLE than Resolve so the right one, maybe people on acescentral can help, idk).

edit: On an another subject, I also tested AsShotNeutral values that I derived from a CR2 I took with my camera on a cdng, thinking that it might be the cause for the +15 tint and just ended up with totally wrong white balance value in Resolve's raw tab, so that's not that but you all probably knew.

21
You're welcome.
You'll understand this better than me:
https://forum.blackmagicdesign.com/viewtopic.php?f=2&t=87576

CaptainHook at Blackmagic Design talking about Gen 1
Quote
Blackmagic Film" for Gen 1 is actually not a transform but passes out sensor space, or sensor RGB. You can basically think of it as "no colour science applied". So you would need a transform from that particular sensors response to the gamut of your choice which you'd need to get from that manufacturer. I believe Digital Bolex for example offered LUTs to transform from their sensor RGB to common gamuts like 709 via a LUT to be used in this workflow with DNGs in Resolve.

Now to he talks about Gen 4 which is a different beast:
Quote
Blackmagic Design Wide Gamut Gen 4 is a common gamut with defined primaries that all our cameras can use (with raw files in Resolve you can choose Gen 4 for all our raw capable cameras). So that is why you can use "Blackmagic Design Broadcast Film" or "Blackmagic Design Pocket 4K Film" colour space/gamut interchangeably as John Paines points out since both of those cameras shipped with Gen 4 - but ideally these options will be consolidated to "Blackmagic Design Wide Gamut Gen 4" for colour space/gamut throughout Resolve.

A member who tried to import Sony shots, Jacob Fenn at the end of the convo
Quote
My issue ended up most likely being that the DNG's weren't relaying the ISO data needed for Resolve to properly debayer to Blackmagic's log curve. Therefore, when I'd apply what I thought was a normalizing transform via a CST node, the result was off because that CST node expects ISO 800. Thanks CaptainHook for help figuring that out.

The ISO data he's refering to isn't what he's talking about really, nor the "debayer to Blackmagic's log curve" he's babbling about. We can totally change the UniqueCameraModel tag for a blackmagic camera and access those ISO settings in the raw tab (spoiler: it's pretty useless for exposure but you can access any blackmagic Gamut/Color Space like the Blackmagic Design Wide Gamut Gen 4 for exemple).

It is probably the BaselineExposure exif tag, I compared with Ursa mini cdng and it can only be this one, but I think I already made this clear.

I agree it is an awful option in MLV App, as there is no technical information about BMDFilm anywhere, like expressions for the log curve, or matrices for the gamut (which seems to be different on every camera that shoots it).

I don't really want to give an opinion but if you don't deem it accurate or reliable well...

CaptainHook talking about clipping, just to hammer the previous argument that using bmdfilm (or any other gammut/colorspace; appart from linear that I can conceive for streamlined SFX workflow) in the raw tab is useless; using rec709 and a CST to your fav log as a first node is as good.
Quote
Yes, DNGs in the metadata provide the matrices and AsShotNeutral tags needed to convert from XYZ to SensorRGB (and back) as supplied by the camera manufacturer so Resolve will be converting from SensorRGB -> XYZ -> Rec709. As for getting an XYZ output option I'm not part of the Resolve team so you would have better luck requesting that direct to them, but you could use CST to go from Rec709 primaries to any other gamut as the output from the RAW decode should not be clipped.

Then maybe I misunderstood.
Also, sorry for the non MLVApp related stuff.

22
Definitely. Even if there's a slight difference between Adobe Camera RAW and MLVApp, let's say the quarter exposure thing, Resolve is still a stop below.
edit: but the fact that some people here doesn't find that result with Resolve is pretty intriguing nonetheless...

23
Well it's like you said Danne, if adobe's DNG is the standard, then passing out the Baseline exposure (+0.26) to the cinemadng exported file gets you closer to said standard.
I totally understand that Blackmagic's Resolve cinemadng rendition isn't the standard, they have added extensions to cinemadng since the ursa mini late 2014 for exemple.
I don't mind the dismissive answers but some comparisons with the exact same settings with another NLE like Premiere instead of ACR are still lacking.
I'm probably not the only one having that issue:
https://forum.blackmagicdesign.com/viewtopic.php?t=45045

In the meantime, I'll be comparing the advantages between cdng in Resolve to tonemapped prores from MLVApp in terms of highlight retention.
I'm still wondering if there's an advantage to get blackmagic's extensions to work on MLVApp's cdng. Something I'll check later.

edit: those blackmagic's extensions, it's just the UniqueCameraModel tag that matters, change it to any blackmagic model and you'll get the full raw tab in Resolve as you would with one of their camera.

24
Atm nobody can tell where the issue really is. Nodoby assumes, you did something wrong.
Care to elaborate? Can you suggest a way to accurately do a comparison between the two?
edit: nvm, it doesn't matter.

25
Yeah I'd assume that it's too complicated to explain in details to non ML devs.
It's hard to find the right balance between usability, user experience and simultaneously satisfy power users.
That's one of the reason why a lot of open source softwares seems clunky for newcomers (but there's other factors, like budget and time constraints, the lack of bug/feature reports, stubborn dev team etc).
edit: I think you did a good job so far. I have my own little lua video presets that are just made off of the 5k anamorphic script (if I need 10 bits with a slight increase in resolution or 12 bits depending of the lighting conditions for exemple).

Pages: [1] 2 3 ... 6