SteadXP is finally out

[Orangenz]

48 minutes ago, HockeyFan12 said:

Just out of curiosity, what is "sensor jarring" and what in the video above looks better to you from the other stabilization solutions? I'm pretty new to stabilization and was considering applying for a gig involving post stabilization, so I'm trying to brush up on the current state of the art. 

Micro vibrations to the sensor cannot be taken out by anything in post. IBIS and gimbals have really decreased this a lot. Steadxp just routes external motion data through onto an audio track through the mic port. Unfortunately andrgl doesn't understand it, or the long history of this company that I have followed. I'm sure you've heard many people say to avoid using electronic stabilisation on cameras. Just don't bother  

[HockeyFan12]

1 hour ago, Orangenz said:

Micro vibrations to the sensor cannot be taken out by anything in post. IBIS and gimbals have really decreased this a lot. Steadxp just routes external motion data through onto an audio track through the mic port. Unfortunately andrgl doesn't understand it, or the long history of this company that I have followed. I'm sure you've heard many people say to avoid using electronic stabilisation on cameras. Just don't bother  

What are micro vibrations and why can't they be reduced in post? I'm not familiar with this phenomenon, but I'm also not used to using smaller cameras. I have seen artifacting when OIS on an iPhone works against a gimbal, for instance. Are micro vibrations a phenomenon limited to smaller cameras? This is the first I've heard that term. Why is it more difficult to reduce these than other rolling shutter artifacts or camera shake artifacts? Does it have to do with the frequency of the motion being above a certain threshold (presumably, 12 shakes per second) or is it related to motion blur obscuring information (which in theory wouldn't be a problem at a high shutter speed)? I'd think having gyroscope data would allow for rolling shutter correction that's superior (at least in terms of camera movement, not figure movement) to any current alternative, but the Nyquist theorem suggests any shakes at >12 oscillations per second would be difficult to solve, regardless of shutter speed. Potentially, if the frequency at which the device (not the camera) records is great enough it could get around these limitations with deconvolution, but that's way out of my league.

What's andrgl?

What post workflows do you recommend for stabilization? To my eye SteadXP is the best in the above video and I don't see it cropping any more than the others, at least in that example. The gig I might apply for is for team that stabilizes in post, rather than in camera, and they shoot with significant look around room (6k for 4k center extraction). Their current workflow as I understand it is one I've used for years and which is very limited (Mocha stabilization then hand animating in curves between the first and last keyframes of the same data to remap the motion with approximately correct parallax), so I'm intrigued by devices like this and, ultimately, reprojection workflows, that would record that stabilization data more precisely and in 3D in order to create a point cloud from which to fill in missing and occluded detail. Of course, I doubt the current software does this, but it's something I've only seen possible in the past with match moving (extremely difficult on the kinds of moves this product is designed for). Furthermore, for match moving generally, the data seems extremely useful as a guide. I don't know much about this thing so maybe I'm excited about the potential of a similar product, not the product itself. 

But to my eye that video looks good and I don't see it cropping substantially more severely than the other software.

To be fair, other than Mocha (which I use most often), I've only really fooled around with Warp Stabilizer and Mercali, and have had better but still pretty poor results with Warp Stabilizer. I'm considering applying for this gig because I want to learn more about stabilization techniques outside the very limited stock packages, so I'm researching stabilization workflows and this caught my eye.

[Orangenz]

1 hour ago, HockeyFan12 said:

What are micro vibrations and why can't they be reduced in post?

Relatively large motions (low frequency) affect the entire image and thus can be processed out. Higher frequency oscillations show up the rolling shutter coming off a sensor and some processing can be done. The highest frequency oscillations corrupt each frame and little can be done to remove this. What camera do you have? 

[HockeyFan12]

11 minutes ago, Orangenz said:

Relatively large motions (low frequency) affect the entire image and thus can be processed out. Higher frequency oscillations show up the rolling shutter coming off a sensor and some processing can be done. The highest frequency oscillations corrupt each frame and little can be done to remove this. What camera do you have? 

That makes sense. If the oscillation is happening at a frequency greater than 24 hz you'd definitely struggle with isolating rolling shutter. I'd think the gyroscope could theoretically be synced with the rolling shutter so it could target different parts of the frame differently, but I doubt that's been implemented or it's even close to precise enough in practice.

For my hobby stuff, I use a C300 (and Ronin with DJI Focus) and iPhone X (and DJI Osmo Mobile not yet working...). The C300 has bad rolling shutter, but I've never noticed anything too crazy with it that can't be fixed in post. The job I might apply for would be 6k Red but require extra stabilization on dolly moves, etc. but that wouldn't be something I'd be on camera department for, so it's just curiosity whether something like this would be useful.

I was interested in this device mostly for a travel kit that's better than the iPhone. (Probably an XC10 or X7S.) Gimbals are cool, but I'm not a good enough operator to eliminate bounce and I wanted something extra to bring things to the next level. Then again, warp stabilizer doesn't do a bad job in with gimbal footage... just not always great.

Would the crop with the SteadXP be worse or better than warp stabilizer? Do you know how it deals with parallax if at all? Adobe's subspace warp is pretty bad in terms of how it distorts footage.

[Orangenz]

1 hour ago, HockeyFan12 said:

That makes sense. If the oscillation is happening at a frequency greater than

Film something and knock the side of the camera. 

[HockeyFan12]

23 hours ago, Orangenz said:

Film something and knock the side of the camera. 

I believe you, I just don't see how that would be a problem under normal circumstances.

Does it crop more than the other software solutions you listed? I feel like I'm missing something. I'll definitely hold off until reviews are in, though.

[jonpais]

Even if it were a viable solution, I'd hold off on purchasing the SteadXP, especially if it's not urgent and your primary reason for getting it is for its small size. virtualGimbal, a microSD to SD adapter that contains a 3-axis gyro sensor, will be coming to market soon enough. The creator says it's for those who don't want to purchase a 'full blown brushless gimbal', even though brushless gimbals have never been so small or affordable as they are today, don't degrade the image in any way, and not only reduce camera shake, but also help make expressive, enlivening camera movement possible, something electronic stabilizers can't do at present. 

[HockeyFan12]

Still curious why there's so much hate for this product. The test video looks great. Leagues better than the virtual gimbal test video.

I'd be using this with an XC10 and a Ronin M most likely. I don't find gimbal stabilization smooth enough, too much translational bounce by far, but the full-sized z axis rigs are huge and unreliable (spring bounce is tuned to a specific walking speed).

That said, I appreciate the feedback and will hold off for now! I think a hybrid solution (gimbal-compatible remote head with gyroscope and pan/tilt data) would be really cool. Maybe some day.

[cantsin]

Again, this cannot work as well as people think. When you shoot video, then normally at 1/50 shutter speed (unless it's a GoPro or cell phone camera in auto mode shooting in bright daylight - but then you end up with stroboscopic motion rendering). If you shoot handheld, camera shake will not only cause a jumpy image framing, but high degrees of motion blur - typically, completely blurred-out frames with jerky movement, or strongly blurred frames with typical unstabilized movement.

No software stabilization, no SteadXP, can remove these damages to the image. You need proper hardware stabilization - through a steadycam, gimbal or good in-camera optical stabilization - to avoid this.

[Orangenz]

5 hours ago, HockeyFan12 said:

I believe you, I just don't see how that would be a problem under normal circumstances.

Does it crop more than the other software solutions you listed? I feel like I'm missing something. I'll definitely hold off until reviews are in, though.

This is not advertised for normal circumstances, rather, extreme circumstances. I believe it crops more, yes. And yes, wait for the reviews of a release version, if it is ever released. 

2 hours ago, HockeyFan12 said:

Still curious why there's so much hate for this product.

A lot is from interaction with the company over many years with great periods of no communication or false release info or status. New people are coming on board and there is literally zero difference in status or videos from many years ago when first put forward and the old guard have given up. 

1 hour ago, cantsin said:

No software stabilization, no SteadXP, can remove these damages to the image. You need proper hardware stabilization - through a steadycam, gimbal or good in-camera optical stabilization - to avoid this.

^^ this

[hmcindie]

7 hours ago, cantsin said:

Again, this cannot work as well as people think. When you shoot video, then normally at 1/50 shutter speed...

1/100 still looks great and is easy to stabilize. Sony does some really good stuff with electronic stabilization, the xperia xz premium phone that I have has very good EIS for video, which only breaks down as the shutter speed goes too low.

So it can work really well. Also your argument is slightly invalid because people have been using Warp Stabilizer a huge amount. It is basically electronic stabilization. We use that a shit ton. Also some optical stabilization systems (for example the Olympus OM1) can use electronic stabilization added on top of the optical.

It's not really about competing with gimbals.

[dhessel]

As someone who does post stabilization all the time I can tell you there are some things you just cannot remove in post. As good as this may be it is not going to ever be as good as a true gimbal. Sure you can use 1/100 shutter to reduce the issues but it looks terrible unless you want it to look like a video game.

Never rely on warp stabilization, I will come back to bite you at one point or another.

[HockeyFan12]

13 hours ago, Orangenz said:

This is not advertised for normal circumstances, rather, extreme circumstances. I believe it crops more, yes. And yes, wait for the reviews of a release version, if it is ever released. 

A lot is from interaction with the company over many years with great periods of no communication or false release info or status. New people are coming on board and there is literally zero difference in status or videos from many years ago when first put forward and the old guard have given up. 

^^ this

I appreciate the heads up on the company not always following through on promises. I'll wait for reviews/a real release.

15 hours ago, cantsin said:

Again, this cannot work as well as people think. When you shoot video, then normally at 1/50 shutter speed (unless it's a GoPro or cell phone camera in auto mode shooting in bright daylight - but then you end up with stroboscopic motion rendering). If you shoot handheld, camera shake will not only cause a jumpy image framing, but high degrees of motion blur - typically, completely blurred-out frames with jerky movement, or strongly blurred frames with typical unstabilized movement.

No software stabilization, no SteadXP, can remove these damages to the image. You need proper hardware stabilization - through a steadycam, gimbal or good in-camera optical stabilization - to avoid this.

I don't entirely agree with this, but I agree that stabilization should be applied to the most stable shot you can get. Of course it's going to degrade the image, but so does literally everything. It's a matter of trade offs. There are a ton of stabilization artifacts in Birdman, which I think was shot on a Steadicam mostly? But they got the image quite stable without ruining it completely, so I like that hybrid approach of stabilizing a relatively stable shot into something smoother.

But it still looked pretty rough at times. That approach can improved upon. Especially as this kind of stuff is pushed further and further into post over time... Already, 6k for 4k extraction and 3.2k or 2k for 1080p extraction seems to be the norm. Pretty soon it's going to be point cloud photography opening up much greater possibilities.

But the Lytro is many years away. The idea of implementing matchmoving to create reprojected geometry seems like the best stopgap, and this device is a step in that direction. Shooting at a high shutter speed and adding motion blur in post isn't the worst idea in the world, either. If you have the camera move information, you're starting with much more accurate vectors.

[dhessel]

Post motion blur can be ok but is not a replacement for the real thing. It can be a fall back solution but shouldn't be your go to solution. There were times in the past at my work where we took the risks and did things like this that were good cheap work around that gave us good results. But eventually it comes back to bite you and all of your tricks and shortcuts fail then you are SOL. 

[HockeyFan12]

42 minutes ago, dhessel said:

Post motion blur can be ok but is not a replacement for the real thing. It can be a fall back solution but shouldn't be your go to solution. There were times in the past at my work where we took the risks and did things like this that were good cheap work around that gave us good results. But eventually it comes back to bite you and all of your tricks and shortcuts fail then you are SOL. 

I'm not so sure. Even basic stuff like RSMB can be the saving grace of claymation, even applied at default values.

Granted my experience was only on ultra-low budget work, but RSMB with motion vectors out of Maya is good enough to replace real (and much slower to render) 3D motion blur. If the SteadXP records motion data accurately enough and the software side as as good as RSMB, the result might be better than the alternatives. Not as good as the real thing, but increasingly what you're seeing on screen isn't the real thing.

[dhessel]

It can work and it can work well but the problem is when it doesn't work there is nothing you can do. It is always better to get it right in camera if you can. SteadyXP seems fine, I had some interest in the data in gather for match moving but there is only so much it can do. For instance at night if you have unsteady footage with jarring motion it might be able to smooth it out but there is nothing it can do about the streaking you will see in lights. Just one such example but it illustrates the problem with relying on less than optimal solutions, they work but sometimes they don't and you won't know until you are back at your workstation trying to make delivery.

[HockeyFan12]

3 minutes ago, dhessel said:

It can work and it can work well but the problem is when it doesn't work there is nothing you can do. It is always better to get it right in camera if you can. SteadyXP seems fine, I had some interest in the data in gather for match moving but there is only so much it can do. For instance at night if you have unsteady footage with jarring motion it might be able to smooth it out but there is nothing it can do about the streaking you will see in lights. Just one such example but it illustrates the problem with relying on less than optimal solutions, they work but sometimes they don't and you won't know until you are back at your workstation trying to make delivery.

Maybe not now (that specifically was a problem in Birdman and a failure of their motion stabilization technique), but with better software, the motion vectors could be applied to more accurate deconvolution, to patching using previous or subsequent frames with accurate tracking data, or even to reprojection with a point cloud.

[Dave Maze]

I just see this technology as an addition to gimbal work. Almost a safety net if your gimbal shot screws up. And that’s perfectly fine.