Scarlet Hybrid Image Stabilization

[Steve Wake]

Originally Posted by Jarred Land in the (now locked) 10 questions thread, May 2010. http://reduser.net/forum/showthread.php?p=604431#post604431

Your correct. No Optical image stabilization, but instead a hybrid stabilization using the sensors in the camera for movement and then shifting the window.. its all metadata so the stabilization correction can be applied, or discarded.

Someone asked me about the fixed focal length in 2/3" broadcast terms... it is F2.4 7.5mm - 58mm. Constant aperture throughout the zoom range.

This was in answer to a question about OIS on the Scarlet Fixed 8x zoom. I had forgotten that RED did foresee use of those motion sensors (feature 22) for stabilization. Of course it will be limited by the motion blur in the individual frames. At some point the stabilized footage will have blur that becomes counterintuitive. But Jarred made it sound so easy in post, it could be very useful and convenient.

Questions:

Is this still planned and will it be done by RED or a third party?

Do other RED cameras have the motion sensors, or is it only a Scarlet Fixed feature?

 

[Michael Dearworth]

Jarred stated that pretty matter-of-factly, so I would be surprised if they took it out. Would be nice to hear from RED about this though.

 

[Trevor Harris]

If its shifting the window... are we going to get black bars on the sides of the image because of this???

 

[Laco Gaal]

If its shifting the window... are we going to get black bars on the sides of the image because of this???

I think of course you'll have to zoom in a bit, as in every way of stabilizing. But compared to the software-only stabilization, this would handle shots, that are hard, or impossible for software stabilization. (for example a shot with a lot of reflections, or a moving subject, or anything distracting in the foreground).
This would be amazing.

As for the motion blur, you can always prepare for it (if you know, you're going to stabilize it, you will need to use a higher shutter speed), or use the HDRx A-track for normal shutter speed, and HDRx X-track with a higher shutter speed, just to be sure.

I like this thing in RED: one great innovation helps the other.

 

[Michael Olsen]

Unless I'm mistaken (and I certainly may be) there is some lookaround area on the 3K sensor even at "full 3K". In that case, there would be no need to push in or lose image to black bars.

 

[Steve Wake]

AFAIK the lookaround area is not recordable. You can set frame lines inside the recorded area in anticipation of the need to stabilize in post. Posts #27 and 28 in the RECON 007 thread kind of clarify the limits. So my interpretation of these posts is there are two ways of thinking about lookaround:

1. There are the extra pixels that one sees around the border of the frame that cannot get recorded, but are there to help you frame the shot. If you are new here, this is, as far as I know, a unique RED feature. RED designs the sensors larger than the maximum recordable frame size and displays the extra pixels. This is the "Lookaround" feature.
2. Then there is the potentially greater lookaround you get by (in Deanan's words) "setting the record area to larger than the intended target format and using inset frame lines for framing."

Being "only" 3K, Scarlet wouldn't have the same wiggle room for stabilization in post as an Epic. For run and gun I would just record the largest area possible, with inset frame lines to help, but then determine the crop for stabilization in post by balancing resolution loss against the need for a steady shot.

 

[Steven Caesare]

AFAIK the lookaround area is not recordable. ...

There are the extra pixels that one sees around the border of the frame that cannot get recorded, but are there to help you frame the shot...

Well... the 4.5K mode that was eventually enabled on the R1 via later firmware actually DID record what had been the left & right lookaround areas. However you then had limited height aspect ratios.

So, it would seem likely that it's not that the lookaround sensor area can't be recorded, but that the overall camera throughput has been engineered to accomodate shooting up to the advertised number of pixels. For the fixed Scarlet that would seem to be 3072x1620=4,976,640 pixels per frame.

So, there's the chance that the image stabilzation systems might take advantage of this. While it would never record greater then the 3072x1620 window resolution, it could shift the entire recorded window area on both the vertical and horizontal axes in response to the motion sensors in the camera.

I have no idea if they ARE doing this.... I just suspect they COULD, as the lookaround area pixels aren't somehow different, and there's norreason they shouldn't be recordable, other than not to exceed the overall system throughput.

-Steve

 

[Michael Olsen]

Thanks for clarifying, Steve. That's exactly what I was thinking. I definitely don't know if that's what they are actually planning on doing, but that's what I thought they were planning on implementing - shifting the recorded area around on the sensor slightly to compensate for slight movement. We'll see, I suppose...

 

[conrad gaunt]

it could shift the entire recorded window area on both the vertical and horizontal axes in response to the motion sensors in the camera.

-Steve

It seems unlikely as Jarred said it could be disabled.. unless by default extra "look around pixels" were always recorded for each resolution.. which also sounds unlikely.. although in theory just enough "extra" resolution could be recorded in reponse to sensor movement.. but then redcode would become much more complicated.. so, I doubt it

 

[paulherrin]

related thread on the 3-axis motion sensor: http://reduser.net/forum/showthread.php?t=54801

i would expect the foundry to have solutions for this in storm, and their other software. hopefully adobe will follow suit, and hopefully final cut will update themselves. if nobody else makes anything by shipping time, i'll write something when i get mine.

 

[Steven Caesare]

It seems unlikely as Jarred said it could be disabled.. unless by default extra "look around pixels" were always recorded for each resolution.. which also sounds unlikely.. although in theory just enough "extra" resolution could be recorded in reponse to sensor movement.. but then redcode would become much more complicated.. so, I doubt it

Hmm, you cerrtainly do have a point there about the fact that it can be "discarded" metadata... so thet does sound like something that could not in fact be a permanent decesion to shift the pixels that are at record at shooting time, becasue it appears you decide to ignore it in post.

So now I'm wondering if it's just sonsor position/acceleration meta ata that is recorded, and the stabilization takes advantage of the 3072x1620 resolution to allow you to extract a smaller stabilized window from it. You'd still have enough resolution to scale down to pristne HD evin if you discarded a few hundred pixels on each axis...

Hmmm...

-sc