Welcome to our community

Be a part of something great, join today!

  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

KOMODO....

Status
Not open for further replies.
I've seen two VFX productions tap into the gyro data thus far. It's in there, but putting RED on top of developing the tools to do more with it is a bit weird. Maybe I'll make a dumber program or something. I did similar for clipJoiner, but I'm a full time filmmaker, it's hard to be tech support on stuff like that :/
 
You can pull the data with REDline and import it into your software.
 
There are a lot of angles (pun fortunate), on capturing positional data during acquisition for a host of post processes. Custom integration using RED's SDK is great if that's your jam, but it seems like an awfully heavy lift for a large segment of the user base.

Komodo is interesting in no small part, because its small - heck the current DSMC2 is very light and compact compared to historical norms. This means more gimbal, handheld and drone shots will seem like a good idea to more people - and the modern style tends more toward an active camera. These are scenarios where a tool in RedCineX - and in the SDK - that supported stabilization in at least 2 axis - likely pitch and roll - could be a real lifesaver.

Note to potential developers: [I'd like a slider, in the -100 to +100 range, 100 being as glass smooth as possible, zero means as shot, no adjustment, -100 is amplification of motion vectors to make a stable shot look as if it was shot wildly.]

I want to note that bringing AI stabilization to the workflow is not just for taming shaky cam. We already do intentional camera movement as a way to "spot" story beats like the audio does. But there are shooting scenarios where that's not going to be possible. What if shot handheld/active style, but in post you could keyframe the stabilization effect to moments/POV changes, etc where the camera reveals a character, or other key moments in a chaotic scene. Even add some active camera in post when desired, especially nice for matching coverage where a static shot would kill the momentum of the scene, etc.

Should RED be responsible for adding features such as these to their free software? That is certainly NOT what I am saying.

What I am suggesting, is that if RED were to add this kind of functionality to their core value proposition, it might be a compelling reason to choose RED over other options. Yes, they'd have to spend money on engineering resources to implement. FWIW, as the disparity in image quality between the low end cameras and the top end continues to narrow, the feature set starts becoming the key differentiator. Other than huge increases in HFR capabilities, I can't think of a better pitch than offering simple/intuitive stabilization in post. [plus, perhaps some engagement with the VFX folks to foster a more accessible framework if they'll buy in] Metadata may not be sexy, but if REDs were the darling of the VFX peeps...

Cheers - #19
 
Just bear in mind that stabilization in post requires forethought. If you stabilize a shaky shot that was shot at 24fps with a normal 180 degree shutter, you'll end up with a bunch of blurred frames. You'd need to shoot it at a much smaller shutter angle and if necessary, add temporal blur in post to match (as well as is possible) a normal 180 degree blur.
 
i is the
You can pull the data with REDline and import it into your software.

Aha. Will try it. Would be easy enough to bring in the raw data in flame and scale to taste.

Just bear in mind that stabilization in post requires forethought. If you stabilize a shaky shot that was shot at 24fps with a normal 180 degree shutter, you'll end up with a bunch of blurred frames. You'd need to shoot it at a much smaller shutter angle and if necessary, add temporal blur in post to match (as well as is possible) a normal 180 degree blur.

All ill cameramoves are not hard shakes. If you use a dampning device like a steady cam for example or a gimbal then you can use gyro data to smooth and level if things are drifting for example.
Also the data is usable when inserting CG. If having your camera on a tripod and panning around you could for example do splitscreens like if you had a fixed camera etc etc. So there is lots of places where this data can be used where the motionblur from the original camera move is a no issue. If the canera and lens is well known you can also counter steer rolling shutter abd motion blur to a very high degree, just take a look at gopro fusion footage before and after post processing and you will understand what I mean. Tha capture can look like a turd and still after processing its silk smooth without rolling shutter ichias.

There are a lot of angles (pun fortunate), on capturing positional data during acquisition for a host of post processes. Custom integration using RED's SDK is great if that's your jam, but it seems like an awfully heavy lift for a large segment of the user base.

Komodo is interesting in no small part, because its small - heck the current DSMC2 is very light and compact compared to historical norms. This means more gimbal, handheld and drone shots will seem like a good idea to more people - and the modern style tends more toward an active camera. These are scenarios where a tool in RedCineX - and in the SDK - that supported stabilization in at least 2 axis - likely pitch and roll - could be a real lifesaver.

Note to potential developers: [I'd like a slider, in the -100 to +100 range, 100 being as glass smooth as possible, zero means as shot, no adjustment, -100 is amplification of motion vectors to make a stable shot look as if it was shot wildly.]

I want to note that bringing AI stabilization to the workflow is not just for taming shaky cam. We already do intentional camera movement as a way to "spot" story beats like the audio does. But there are shooting scenarios where that's not going to be possible. What if shot handheld/active style, but in post you could keyframe the stabilization effect to moments/POV changes, etc where the camera reveals a character, or other key moments in a chaotic scene. Even add some active camera in post when desired, especially nice for matching coverage where a static shot would kill the momentum of the scene, etc.

Should RED be responsible for adding features such as these to their free software? That is certainly NOT what I am saying.

What I am suggesting, is that if RED were to add this kind of functionality to their core value proposition, it might be a compelling reason to choose RED over other options. Yes, they'd have to spend money on engineering resources to implement. FWIW, as the disparity in image quality between the low end cameras and the top end continues to narrow, the feature set starts becoming the key differentiator. Other than huge increases in HFR capabilities, I can't think of a better pitch than offering simple/intuitive stabilization in post. [plus, perhaps some engagement with the VFX folks to foster a more accessible framework if they'll buy in] Metadata may not be sexy, but if REDs were the darling of the VFX peeps...

Cheers - #19

Again, correcting tilt is a whole different level than correcting roll. To correct tilt you need to know the lens and the distortion and likly your camera is not in the nodal point so there will be paralax change that can not be accounted for. So thats a much, much bigger thing to tap into than roll.
Roll you can eliminate without knowing anything about the lens.

All needed to stabilze roll is: Negating the gyro values for the roll axis. Scale and offset those values ( which can all be done in any half deasent chanel editor.) So all you need for roll is basically a raw string of numbers for each frame that imports as a raw curve. Will try to extract from a shot in redline as soon as I get to the office.
 
It would still be a win if you could easy extract the tilt axis. Say you have a static tripod shot that you like to do a projection of to geometri. Then it would be nice to have the exact tilt angle in the meta data. Or you already have an backplate for a shot and you are going to shoot an foreground clip in green screen to match that background.
 
There are clearly limitations on what can and cannot be addressed with a particular technique. SteadXP suggests there's potential for embedding positional data in metadata to enhance stabilization coherence. If a first tier camera company like RED made it a priority, I'd be very interested in what might be possible. Thanks to smart phones, game controllers, etc the tiny low power chips needed should be both capable and cheap. I think it could be a meaningful value add.

Cheers - #19
 
It would still be a win if you could easy extract the tilt axis. Say you have a static tripod shot that you like to do a projection of to geometri. Then it would be nice to have the exact tilt angle in the meta data. Or you already have an backplate for a shot and you are going to shoot an foreground clip in green screen to match that background.

Yes but if you dont know the lens things will not move trough the frame properly. Think if you got a 200mm lens and tilt a few degrees things will move very linear and fast trough the frame with little distortion. If you got a 8mm lens it will be the opposite.


But yes if set up in a 3D scene correctly with the right size backplane lens, and distortion etc then it works very well.

I looked now, now wonder why I never seen this redline app... its apperently now only windows a linux and windows app.

From redcine manual:


"You can launch REDline in Terminal on Command Prompt on Windows, or shell on Linux®. For Windows, REDline is located in the default installation location of the REDCINE-X PRO® suite of applications. For Linux, you must download the standalone REDline Linux (Beta) software."

Hackingtoch is a wonderful thing.

Rebooting into windows...

:)

Edit: Not so easy... Who wrote the app, anybody knows?
 
I suspect we ought to start a new topic on metadata?

I wasn't aware of RedLine and will take a look. I can see in the latest SDK in the metadata.h mentions of tilt, pan and roll. However aside from RedLine just mentioned i'm not aware of any vfx software reading that meta data (unless things have changed recently).

I'm going to bounce a question to the engineers at foundry for this one - as it shouldn't be too tough to pull the data... if it's there.

Has anyone extracted this via RedLine to see whether the data is there and working?

EDIT: In the latest Nuke i do see the rotation and acceleration data but these aren't updating each frame. So i've bounced this across to see if it's a bug or the data....

cheers
Paul
 
I've seen two VFX productions tap into the gyro data thus far. It's in there, but putting RED on top of developing the tools to do more with it is a bit weird. Maybe I'll make a dumber program or something. I did similar for clipJoiner, but I'm a full time filmmaker, it's hard to be tech support on stuff like that :/

No one demands that Red put this feature in Redcine or SDK for third-party applications, but no one did so with the numerous features we already got from Red dealing with R3D RAW data. As many here mention, there are VFX teams pulling that data from the metadata and work with it, but that's just it; a slow, complicated process to get to the end result. For camera tracking, we, of course, need more metadata, but if our cameras do SteadXP-like gyro recording per frame, then the whole point would be that we have this as a RAW setting feature.

Imagine just pulling in R3D files into an NLE, Redcine-X or finishing application like Resolve and just slide a "stabilize smoothness" slider in the RAW settings from "as-shot" to "floating".

I've worked with SteadXP and the result is mind-blowing but too cumbersome, it's the ease of use in just having it as a RAW setting that makes this a powerful setting, and probably one of the most powerful and understated settings of most cameras out there. If the metadata is already there it wouldn't be much trouble putting in this feature, not with the track record of Red adding features to the R3D workflow. Getting butter-smooth gimbal-like shots from handheld, or stabilize a rough dolly to Fincheresque quality in a push of a button has tremendous potential for filmmakers, especially those without an expensive VFX team.
 
Same with the foundry, just heard back. So hopefully can use that data and see what it looks like for real in the not too distant future....l

I sent a new feature request to autodesk sounded like they will pick up the ball.

If it gets implemented in an easy-to-use way (hardware-assisted post-stabilization that doesn't take forever to analyze footage), this would obviously be a welcomed feature. If it were that simple (e.g. a slider in RCXp et al), I'd wonder why it wasn't taken advantage of a lot sooner.

...Having that metadata in there since ~2012/DSMC1, but only now getting a real-world use from it because other companies (SteadyXP, C500mk2's internal-but-non-IB IS, etc) have started doing it, is bittersweet (or 'dropping the ball'). Especially since RED cameras are frequently used for VFX.
 
Worth an ask

Worth an ask

1) Perhaps RED expected 3rd parties would use the SDK to develop tools well suited to VFX needs

2) Perhaps the processing muscle required makes it painfully slow - or at least it did in 2012

3) RED was focused on making cameras, developing sensor tech, implementing IPP2, etc

4) Motion blur that constrains the utility of even the best stabilization tool

IAC, if they consider it important enough, our favorite "computer with a lens" might be able to take stabilization to the next level.

Cheers - #19
 
Interesting article about Steven Soderbergh in Filmmaker Magazine. He mentions shooting on Komodo.

https://filmmakermagazine.com/10831...at-llcs-and-systemic-corruption/#.XYVtJcplCfA

The bomb has dropped on the sensor size :)



Soderbergh: What’s interesting is the film that I made since The Laundromat with Meryl that we just wrapped, I was using the new Red Komodo, which is the best of both. It’s extremely small, but it’s got a full-sized sensor. You can strip it down to absolutely nothing and it’s truly quite small and light. The lens is bigger and heavier than the body. Or you can very easily—if you need a little bit of mass because of the shot you’re doing, whether it be handheld or whatever—put some gak on it and give it a little more weight and turn it into that tool. We were the first people to get it. These bodies were hot off the bench. I was very, very pleased with what it allowed me to do. It was small enough for me to be able to put it anywhere I wanted very quickly but big enough to give me the aesthetics and the practicality that a “normal” camera would give me. It was pretty exciting.

Filmmaker: Are you going to use it going forward?

Soderbergh: Yeah, I mean, I don’t know why I wouldn’t. I guess my question for them is: are people supposed to use this now? I have three Monstro bodies. I’m wondering what I do with those. Rent them out, I guess.​
 
Status
Not open for further replies.
Back
Top