Welcome to our community

Be a part of something great, join today!

Log in
  • 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 :)

HDRx with More Normal Motion Blur simulation

Floris Liesker

Floris Liesker

Well-known member
Joined
Dec 29, 2006
Messages
170
Reaction score
0
Points
0
Website
vimeo.com
I've set up an After Effects project which reduces/eliminates the motion artifacts found in HDRx footage. You can view it here on vimeo.

I've done it quite roughly (from the H.264 Jim put online) but it seems to be working well.
It basically extracts the highlights of the footage, (above RGB 160,150,140), analyzes the movement in there and adds motion blur to it. Then it blends them with the original shadows again.
The darker part of the image is untouched.

Original%20footage.png


MNMB%20simulation.png


The effect is most visible when the LED hand sign sweeps through the frame.
Also watch the lights above the entry of the building left of the hand sign.
In the original film you can quite clearly see a greyish clipping of the long exposure and a fast-shutter version behind it.
In the motion blur version it looks more like a normal exposure.
I think this effect will work just as well with a 6+ setting (I think this is shot at +3) giving you a highlight headroom of 6 stops on top of the 12+ stops you already have!

Funnily enough: when in motion it's very hard to even distinguish the two versions!
I don't know why that is, must be something with the way the eye/brain sees movement.

I guess I could live with HDRx without any render-heavy MNMB added, at least by the looks of the moving footage.
The stills do look a lot better with it though.
Very curious about new HDRx footage. Can't wait for the new cameras to come out.
 
I got an idea: I read Adam Wilt's blog about this HDRx and I remember him saying that this adding of motion blur was something that could never be done in-camera, and my After Effect render times sort of confirm that.
However.
Many modern flatscreen TV's add in-between frames to make motion more smooth. Watching a film in that mode is like degrading it to a soap, in my opinion.
But nevertheless, a mid-priced flatscreen tv is capable of creating new interpolated frames based on motion estimation IN REAL TIME. I guess some sort of single-purposed hardware is doing the math for that.
Those algorithms could easily, instead of creating 8 new frames, blend those 8 together (the 180 degree samples, mind you) and create a smooth motion blur. In real time.
I suppose it wouldn't be doing as good a job as the Foundry software is doing, but hey, this is real time! And I wouldn't be surprised if it was cheap too, being mass-made.
Oh, the mass made stuff isn't 5K res, I forgot. Oh well.
Maybe a 1080p motion blur in the EasyHDR preset would be looking nice enough. In celluloid film the highlights tend to be softer anyway, I think it would look fine.
Graeme, what do you think? Any room left in the camera body for another board?
 
What process/plugins did you use to add the blur?
 
I used CC time warp, which has a built in motion blur. It is very slow though, Reelsmart Motion Blur is supposed to be lightning fast in comparison, but I don't have that one.
There is a lot of fiddling with the gamma's involved to make it behave like linear light, or as close as possible.
Best would be to have the highlights separately and doing the motion blur in there before it gets mixed in with the shadows. It's kinda hard to separate them nicely afterwards.
 
Cool! I actually LOVE the look of Magic Motion™ but it's always good to have options. Thanks a lot Floris.
 
Shane Betts said:
Cool! I actually LOVE the look of Magic Motion™ but it's always good to have options. Thanks a lot Floris.
Click to expand...

The idea of relying on a computer to guess subframe motion for me is a repulsive waste of time. Not a fucking chance. I want to shoot what happens in front of the camera, not what a computer guesses might have happened.
 
Ruairi Robinson said:
The idea of relying on a computer to guess subframe motion for me is a repulsive waste of time. Not a fucking chance. I want to shoot what happens in front of the camera, not what a computer guesses might have happened.
Click to expand...

Not sure whether you're bitching about Magic Motion™ or Floris's technique here but might I say four things. First, either way, you have the choice, second, Magic Motion™ doesn't invent or guess anything, it just takes two shots of what's in front of the lens at two different exposure times, times that you have control over at the time of shooting and combines them, third, Floris is just offering up another way of treating your footage if you want the option (something he's taken the time to work on and has now kindly posted for the rest of us out of the goodness of his heart) and last, why the attitude? Unlax! :chillpill:
 
Ruairi Robinson said:
The idea of relying on a computer to guess subframe motion for me is a repulsive waste of time. Not a fucking chance. I want to shoot what happens in front of the camera, not what a computer guesses might have happened.
Click to expand...

I didn't take into account that relying on a computer to guess subframe motion is a repulsive waste of time, due to Not A Fucking Chance™.
I therefore added the option to bypass the motion blur.
This prohibits the computer effectively from guessing what might have happened in front of the camera.
The process does take several hours per frame to render but the end result is indistinguishable from the original.

Original:
Original%20frame.png


With motion blur (NAFC™):
MNMB%20not%20a%20fucking%20chance.png


Bypassed motion blur (AFC™):
MNMB%20with%20optional%20bypassed%20motion%20blur.png


Hope this helps,
 
Ruairi Robinson said:
The idea of relying on a computer to guess subframe motion for me is a repulsive waste of time. Not a fucking chance. I want to shoot what happens in front of the camera, not what a computer guesses might have happened.
Click to expand...

So you've never done a sky replacement?
 
Gavin Greenwalt said:
So you've never done a sky replacement?
Click to expand...

If I do a sky replacement, I'm using a computer to match the motion characteristics of the live action plate as precisely as possible.

As I have done numerous times. More times than I'd prefer. I have been to roto hell and lived to tell the tale, and I can't say it did not leave scars.

I just think having a computer interpolate this stuff seems to be going down the wrong road. How does it handle reflections? transparency? Why even bother, when you can capture this stuff for real, perfectly, in camera? I object to this in the same way I object to those gimmicky 200hz mega smooth interpolated motion modes on modern TVs that make Lord of the rings look like a soap opera. Even on shots with no sky replacements.

Maybe I am misunderstanding the way it works though - I have been led to believe that this "more normal motion blur" method requires a motion vector style interpolation of the two exposures, meaning for certain types of motion (fields of grass, reflections etc) this method is pretty much guaranteed to screw up. They just haven't field tested these kind of shots yet. Or maybe they have and it works great, and I'm wrong. I'd be happy with that outcome too.

R.
 
Ruairi Robinson said:
Why even bother, when you can capture this stuff for real, perfectly, in camera?
Click to expand...

Well, that's just the problem. You cannot capture 18 stops of DR for real, perfectly, in camera.
Not without some sort of sacrifice, in this case uneven motion blur from shadows to highlights.
 
Ruairi Robinson said:
Maybe I am misunderstanding the way it works though - I have been led to believe that this "more normal motion blur" method requires a motion vector style interpolation of the two exposures, meaning for certain types of motion (fields of grass, reflections etc) this method is pretty much guaranteed to screw up. They just haven't field tested these kind of shots yet. Or maybe they have and it works great, and I'm wrong. I'd be happy with that outcome too.

R.
Click to expand...

Agreed. In fact I just took a world of flack for pointing these things out in an earlier thread. (Semi-transparencies, repeating textures, reflections, parallax, rotating objects etc...).

On the other hand if it's just two people talking and it's 10 pixels of motion blur without any parallax I'm more than happy to let the computer guess. I would imagine that one would be extremely hard pressed to discern the vector blur from real motion blur on small scales.

And remember the only plate that will be getting the vector blur is the HDRx™ track used to bring detail back into the blown out pixels. If 90% of your scene is exposed properly you don't need any vector blur on that portion of your frame (aka your field of impossible to track grass). On the other hand a cloud in the sky that's sitting all by itself off to the side is a relatively static and parallax free object which you could get some extra dynamic range out of and probably isn't even going to be noticed by the audience nor subject to any scrutiny.


Rob Delafonte said:
I know what you're saying, but I think you need to go to Ruairi's website before being that much of a smart aleck...
Click to expand...

Sorry, I should have added one of these: " ;) " I just was poking fun because I know Rauri is a very talented vfx artist in addition to director and vfx is full of guess work. Any time you track a plate you're guessing the sub frame motion. I don't see the difference between the motion blur in a sky replacement (derived from optical tracking) and getting your motion vectors from optical flow. (Unless there is a decent amount of parallax in which case Rauri is right, you're pretty much screwed every time.) On the other hand a nodal pan across a mostly static scene I can't see getting noticeably screwed up and I doubt anyone would ever notice.
 
Ruairi Robinson said:
I have been led to believe that this "more normal motion blur" method requires a motion vector style interpolation of the two exposures, meaning for certain types of motion (fields of grass, reflections etc) this method is pretty much guaranteed to screw up. They just haven't field tested these kind of shots yet.
Click to expand...

In the example I put on vimeo actually some stuff goes 'wrong'. The cc timewarp has trouble analyzing the light pattern in the left of the screen, it's a repeating pattern, so it doesn't see the motion right. Not unlike our brain, sometimes. Still it's looking fine, though, it just looks a little like MagicMotion without motion blur, which already doesn't look too bad.

As Gavin points out, it's only the highlight track that gets the motion blur, and only on the spots where the shadow track 'allows' the highlights to be added; if the shadow part on that spot isn't blown out it prevents the highlight blur to appear.

I've come up with another way to add motion blur without even analyzing motion by comparing frames and interpolating them. It's unusual and not finished but I think I'm on to something. More on that soon.
 
You must log in or register to reply here.
Back
Top