What's wrong with my gyro stabilization?

Hi all,
I was super excited to get my hands on the new firmware and put the gyro stabilization through its paces. I've seen several demos on YouTube where the footage is coming out smooth as butter, but unfortunately that hasn't been my experience. When I apply gyro stabilization in Resolve, I get a weird shaking/pulsing effect in my footage and overall it's much worse than using the software "perspective" algorithm to stabilize. I shot a short clip so you can see what's happening:

Any idea what's going on here? Here's all the details:

- Pocket Cinema Camera 4K, Firmware 7.9
- DaVinci Resolve 18 Beta 6
- Motion sensor calibrated before shoot on a flat surface
- Meike 16mm MFT lens (native MFT mount)
- 16mm focal length entered into the metadata on the slate page
- Lens is fully manual, no OIS (option greyed out in camera)
- 30FPS, 45° shutter
- Stabilization in "Camera Gyro" mode, applied at 1.0 strength w/zoom

Update: I changed the focal length in the metadata/slate to 32mm to see if that would make a difference (16mm MFT is 32mm in FF), but that seemed to make it even worse. Here's a quick video comparing perspective stabilization (left) to gyro stabilization (right):

As you can see the gyro data is definitely being read by Resolve, it's just being implemented very poorly for some reason. I'm not sure if this is a software error, or if there's something wrong with my camera.

Set your shutter between 75 to 45 and re- test.

If you need put focal length, but not calculate crop, focal length give you perspective proportion, if you crop not change this.
I saw your videos and … I think you should ask too much, gyro give a math exact correction from camera motion, but you roll camera, change the inclination and this isn’t possibile to correct.
Gyro give you precise correction but not magic



Inviato dal mio iPhone utilizzando Tapatalk

Set your shutter between 75 to 45 and re- test.

My shutter is set to 45°, I've also tried at 90° and there's no real difference. Shutter speed only affects blurring artifacts in the footage, it shouldn't affect the stabilization itself, right?

If you need put focal length, but not calculate crop, focal length give you perspective proportion, if you crop not change this.
I saw your videos and … I think you should ask too much, gyro give a math exact correction from camera motion, but you roll camera, change the inclination and this isn’t possibile to correct.
Gyro give you precise correction but not magic

I really don't think I'm asking too much here; I'm just walking around my yard, trying to correct a little handheld shake. This guy is literally running alongside his subject and the gyro is having no trouble smoothing it out (skip to 3:20 for that):

This definitely seems like a hardware or software issue on my end, I'm getting artifacts in my footage that I'm not seeing anywhere else in demonstrations.

His unstabilized footage (unlike yours) is unusually smooth and steady to begin with, which certainly helps. In your split screen video, it looked like the other half wasn't stabilized at all. And it's not cropped compared to the perspective stabilization.

Try mounting the camera on a tripod and carrying that. And make an effort to keep the camera locked on one plane and maintain smooth continuous motion. Operating in a haphazard way ("just walking around my yard") is not going to produce smooth footage.

Hi Mineralsound, the link that you post you can see two different things, first many thing not tilt camera like yours, and second (important) seems shooted at more frame speeds that mean you can double gyro data recorded.
I just played a bit with gyro shooting and if you grab camera with handle, have a right balanced camera / lens to avoid tilt pan roll exaggerate motion you can stabilise better than original option. If you tilt (change angle of camera) you do alteration on parallax that mean is very difficult or impossible to correct.
Try to shoot again and try to keep camera more aligned to horizon and I’m sure it will be stabilised better
Often it’s a matter of where you put the weight center of camera.


Inviato dal mio iPhone utilizzando Tapatalk

His unstabilized footage (unlike yours) is unusually smooth and steady to begin with, which certainly helps.

Are you sure about that? At 3:20 he's running with his subject and the camera is all over the place, and the gyro is having no trouble stabilizing that.

In your split screen video, it looked like the other half wasn't stabilized at all. And it's not cropped compared to the perspective stabilization.

That's exactly what I'm trying to demonstrate! The image on the right is with gyro stabilization at 1.0 strength and it doesn't look stabilized at all. You can see the left side of the frame swing around at 0:23 if you want proof that it's engaged. I unclicked "zoom" on both clips so neither of them are cropped in (did that so the framing is the same).

Hi Mineralsound, the link that you post you can see two different things, first many thing not tilt camera like yours, and second (important) seems shooted at more frame speeds that mean you can double gyro data recorded.
I just played a bit with gyro shooting and if you grab camera with handle, have a right balanced camera / lens to avoid tilt pan roll exaggerate motion you can stabilise better than original option. If you tilt (change angle of camera) you do alteration on parallax that mean is very difficult or impossible to correct.
Try to shoot again and try to keep camera more aligned to horizon and I’m sure it will be stabilised better

It's possible I just had unrealistic expectations for this, but it's just that my footage seems so much worse than everyone else's. And I've seen gyro stabilization used to smooth out much shakier footage than this; on the Gyroflow website (https://gyroflow.xyz/) you can see it being used on helmet cams while trail running down a mountain, no attention being paid to rotation axes there!

In addition, the shaking artifact is present in my footage even when I'm just walking forward in a straight line, which shouldn't trigger any parallax effects. Is it possible that lens distortion may be playing a role here? This is a very wide lens with some definite image curvature.

Hi Mineralsound, the link that you post you can see two different things, first many thing not tilt camera like yours, and second (important) seems shooted at more frame speeds that mean you can double gyro data recorded.
I just played a bit with gyro shooting and if you grab camera with handle, have a right balanced camera / lens to avoid tilt pan roll exaggerate motion you can stabilise better than original option. If you tilt (change angle of camera) you do alteration on parallax that mean is very difficult or impossible to correct.
Try to shoot again and try to keep camera more aligned to horizon and I’m sure it will be stabilised better

It's possible I just had unrealistic expectations for this, but it's just that my footage seems so much worse than everyone else's. And I've seen gyro stabilization used to smooth out much shakier footage than this; on the Gyroflow website (https://gyroflow.xyz/) you can see it being used on helmet cams while trail running down a mountain, no attention being paid to rotation axes there!

In addition, the shaking artifact is present in my footage even when I'm just walking forward in a straight line, which shouldn't trigger any parallax effects. Is it possible that lens distortion may be playing a role here? This is a very wide lens with some definite image curvature.

May be that lens distortion give you more artifacts, I tested with 18-35 sigma which is very low distorted image. Today I do more test, I will let you know more


Inviato dal mio iPhone utilizzando Tapatalk

His unstabilized footage (unlike yours) is unusually smooth and steady to begin with, which certainly helps.

Are you sure about that? At 3:20 he's running with his subject and the camera is all over the place, and the gyro is having no trouble stabilizing that.

If you try running with a hand-held camera, you'll see how much worse it can be. His footage is unusually stable for that kind of movement. (But see proviso below.).

If you want to see, pre-Steadicam, what running with a camera actually looks like, try this, from about 3:24 for the impatient:

(Of course, this camera work serves the drama.)

In your split screen video, it looked like the other half wasn't stabilized at all. And it's not cropped compared to the perspective stabilization.

That's exactly what I'm trying to demonstrate!

My point was, comparing stabilized footage to unstabilized footage doesn't help answer the question you're asking here. It's your footage and you apparently failed to stabilize it. Only you can know what might have gone wrong.

The other thing to bear in mind is, a youtube which doesn't give you access to the original footage proves nothing. Unless you have reason to trust the source (I wouldn't know), there's no way to evaluate what you're seeing. Youtube gimbal reviews (to use another example) are commonly full of misrepresentations and apparently intentional deception, in the apparent search for subscribers and hits.

My point was, comparing stabilized footage to unstabilized footage doesn't help answer the question you're asking here. It's your footage and you apparently failed to stabilize it. Only you can know what might have gone wrong.

I don't know how much more clear I can be; both sides of that video are stabilized. On the left side, the clip has been stabilized using the "Perspective" algorithm with strength set to 1.0. On the right side, the clip has been stabilized using the "Camera Gyro" setting with strength set to 1.0. The footage on the left looks smooth, the footage on the right looks shaky. The point I am making is that the gyro stabilization is not working despite being engaged. Are we at least on the same page there?

Gyro stabilization has been demonstrated to be better than software stabilization algorithms, which is why I included the screen on the left to show that in my case, it is not better. And I do not know what went wrong, which is why I'm posting here.

It's your footage and you apparently failed to stabilize it.

Welcome to the conversation, that's what dude has been saying from the beginning.

Only you can know what might have gone wrong.

Super helpful. What do you think you contribute?

Good Luck

Gyro stabilization is not a gimbal or a steadicam. It just makes micro adjustments to remove the jitters and it does a fantastic job of that. You do have to get it right in camera, gyro just makes right more right.

Gyro stabilization is not a gimbal or a steadicam. It just makes micro adjustments to remove the jitters and it does a fantastic job of that. You do have to get it right in camera, gyro just makes right more right.

No, it is better than that. See this comparison, with the gyro stabilization done right and which includes walking with the camera and 200mm long telephoto static shots:

with the gyro stabilization done right and which includes walking with the camera

That's probably better than most people could expect of a gimbal, but the empty space around the camera mitigates the motion artifacts. Gimbal reviewers love open fields, for the same reason. If you ask me, all this technology solves problems no one at this level really has, but never mind....

As for that other issue, unlike Howard (Good Luck to you too, Howie Roll!), I'm not a psychic. I don't know why posted unstabilized footage wasn't stabilized -- where, as in this case, it appears the process failed entirely, rather than being unsatisfactory. The usual explanation would be user error, but I leave that to Howard as the more insightful and telepathic contributor.

Put in the actual Focal Length of the lens and not the crop factor focal length.

[CaptainHook]

Hi, please try 7.9.1 which has improved motion sensor data accuracy for the Pocket 4K (puts it inline with the Pocket 6K) along with other fixes/improvements.

That was quick! Thanks CaptainHook.

BMPCC4K 4K 2.4:1 75fps has returned.

Am I correctly assuming that it doesn't really work with non-electronic lenses?

I tried to gyro-stabilise at 80mm on a 6K both with a cheap Canon lens which was communicating it's focal length and with a nice classic glass of the same focal length. Even if I input that value in the metadata, it looks awful after trying to stabilise.

The electronic lens got stabilised remarkably well for that long lens, though.

I only have manual lenses, but I spent yesterday testing the gyro stabilization in various ways holding the camera grips either in my hands or by holding the camera arm that is balanced fairly well. Without stabilization the clips look like a madman was filming that can drive the viewer to madness; but, after gyro stabilization, it looks like someone stalking their prey in a slasher flic.

Better, but only if I want to shoot a film of a stalker that horrifies the viewer rather than makes the viewer seasick.

Still better than the warped reality jiggly jello look of the surroundings with Perspective stabilization that only suits a Time Traveller’s trip.

I only have manual lenses, but I spent yesterday testing the gyro stabilization in various ways holding the camera grips either in my hands or by holding the camera arm that is balanced fairly well. Without stabilization the clips look like a madman was filming that can drive the viewer to madness; but, after gyro stabilization, it looks like someone stalking their prey in a slasher flic.

Better, but only if I want to shoot a film of a stalker that horrifies the viewer rather than makes the viewer seasick.

Still better than the warped reality jiggly jello look of the surroundings with Perspective stabilization that only suits a Time Traveller’s trip.

I only have manual lenses, but I spent yesterday testing the gyro stabilization in various ways holding the camera grips either in my hands or by holding the camera arm that is balanced fairly well. Without stabilization the clips look like a madman was filming that can drive the viewer to madness; but, after gyro stabilization, it looks like someone stalking their prey in a slasher flic.

Better, but only if I want to shoot a film of a stalker that horrifies the viewer rather than makes the viewer seasick.

Still better than the warped reality jiggly jello look of the surroundings with Perspective stabilization that only suits a Time Traveller’s trip.

Deleted duplicate.

I tried to gyro-stabilise at 80mm on a 6K both with a cheap Canon lens which was communicating it's focal length and with a nice classic glass of the same focal length. Even if I input that value in the metadata, it looks awful after trying to stabilise.

80mm is pushing it, for any movement. You really want as wide as possible. Also it's not clear that manually inputting the focal length "works". Did you experiment with other values? Here's guessing it's not read, regardless.

Yes, I wanted to stress it. But with a 35mm it's just the same. I love my vintage lenses, sigh.

OK, some more observations: entering the correct focal length for lenses with no electronic connection works sometimes, but not reliably. I can't see a clear pattern yet.

Am I correctly assuming that it doesn't really work with non-electronic lenses?

I tried to gyro-stabilise at 80mm on a 6K both with a cheap Canon lens which was communicating it's focal length and with a nice classic glass of the same focal length. Even if I input that value in the metadata, it looks awful after trying to stabilise.

The electronic lens got stabilised remarkably well for that long lens, though.

Put in the actual Focal Length of the lens and not the crop factor focal length.

I know what you mean by Focal Length of course, but what is the label in the metadata in DaVinci Resolve? Is it Lens Number? That’s the closest thing I’ve found. There’s Focal Plane (mm) or something like that but is that Focal Length?

Yes, that "Focal Point (mm)" entry under "Camera" is the focal length, as is clearly visible if you use a lens with electronics. I don't get it why it's labeled that way.

Of course, I always write the 'true' focal length into that field – since it never changes and DR can read the sensor size for their cameras anyway.
I'm not a fan of that stupid "my lens then becomes a XX mm lens" notion anyway. It's only misleading.

To get the focal length setting to work on manual lenses, you'd probably have to be able edit it into the braw metadata directly. Putting it in Resolve's metadata template doesn't write it to the file, and gryo stabilization probably doesn't see it.

The sidecar might be another way, but that only seems to record the raw panel parameters.

Would be nice to get an official statement from BM. There's not much information at all – the keyword 'gyro' doesn't generate a single hit on those 4,000+ pages…

Yes, that "Focal Point (mm)" entry under "Camera" is the focal length, as is clearly visible if you use a lens with electronics. I don't get it why it's labeled that way...

Thanks for the clarification on that item which apparently everyone else with manual lenses understood.

To get the focal length setting to work on manual lenses, you'd probably have to be able edit it into the braw metadata directly. Putting it in Resolve's metadata template doesn't write it to the file...

Thanks, John.

We were advised previously, if memory serves, to enter the focal length to increase the accuracy of gyro stabilization.

I’ll enter the focal length via the camera slate and retest later. I don’t think I entered focal length that way for my test. If I’m on a shoot with the BMPCC4K, I should enter prime lens focal lengths.

However with a zoom lens varying focal length during a single shot, I’ve no idea how gyro stabilization will cope. Easy enough to test though so I’ll do that with the Tokina 11-20mm manual zoom.

I updated the camera slate and then shot some tests using the SLR Magic APO 32mm with the 1.33x-65 Anamorphot. I defined the Focal Point as 32mm. That worked well in terms of allowing camera movement but no handheld jitters as I walked behind the camera taking small steps. Not Hollywood qualified but maybe in ten years, eh?

For good results, try your Tokina at 11mm.

Forget small steps, that will create more perturbations..... There's a technique for it, "rolling" from the balls of the feet with the knees bent, but it's exhausting, even for 18-year olds. Have you tried mounting the camera on something or other (like a small tripod) and carrying that, horizontally (or rather 45 degrees)? The best thing might be a light two handled device. They make them for gimbals; there might be some way to rig the camera directly on one. You hold the device extended, away from the body, again raised at about a 45 degree angle.

Thanks very much, John.

With the Tokina 11-20mm secured by lens support on rails, it’s much lighter. I tested at 11mm, 20mm, and zooming while walking far to near (from 20mm to 11mm) and then from near walking backwards to far (from 11mm to 20mm). So those two zoom shots are trying those famous dolly plus zoom shots except I’m walking. Done testing but need to get dinner ready… results might wait until tomorrow.

Zooming might be quite a challenge. If I compare it to Gyroflow, which even has a module to generate lens profiles, it's remarkable what the software achieves with just the right focal length.

Does Gyroflow support BRAW on the BMPCC4K and BMPCC6K as of 7.9 firmware?

Well it was interesting to test gyro stabilization yesterday with the Tokina 11-20mm handheld. Results shooting at 11mm and 20mm still show I’m not ready to use this for a client shoot, but with the advice offered I’ll get better with practice. What was more interesting were the handheld tests zooming by hand from 20mm to 11mm walking forwards and from 11mm to 20mm walking backwards.

Considering it was the first time trying such a stunt, the results were surprisingly decent. I walk backwards smoother than forwards! In both directions it looks like I’m hardly moving except you can see a hint of the perspective changes as the focal lengths change. But the subject stays the same size and fairly stable and in frame with the gyro stabilization applied. Too bad my subject was a window and not a person with a pipe and a prominent proboscis.

From further tests I conclude that you have to define the correct focal length in the camera's slate for manual lenses.
Correcting this value in DR doesn't seem to work. As usual, discipline when shooting is needed

So, a stunt like yours will only work with a lens that's electronically coupled.

You’re probably right about needing a continuous feedback to the camera for the focal length changes as you zoom during a shot.

Something like this Olympus 12-40 f/2.8 Pro mark II with autofocus and no stabilization:

https://www.bhphotovideo.com/c/product/1688419-REG

Only drawback though is it’s not really parfocal, is it?

Are there any electronic, parfocal, constant aperture, wide to telephoto, autofocus lenses that will communicate focal length to the BMPCC4K or BMPCC6K G2? I’m holding my breath!

I’m holding my breath!

Now we are two

Just add Selena Gomez and we can start vlogging “Only Stunt Shots in the Building.”

[CaptainHook]

The sidecar might be another way, but that only seems to record the raw panel parameters.

You can add the lens focal length to the sidecar and it will work, but its probably not very practical at this stage since its per frame metadata you would need to at least add the timecode of the first frame of the clip to the sidecar when adding the new focal length metadata. Also, see below:

However with a zoom lens varying focal length during a single shot, I’ve no idea how gyro stabilization will cope.

Resolve does not currently support changing the focal length during the clip with the gyro stabilisation, it will use whatever the value of the first frame is for the entirety of the clip so you may not get results you expect.

Does Gyroflow support BRAW on the BMPCC4K and BMPCC6K as of 7.9 firmware?

In the latest dev builds, yes:
https://gyroflow.xyz/devbuild/?autodownload
They cannot decode .braw files directly though as they don't use the Blackmagic RAW SDK so you need to transcode files first for video, but gyroflow can read a .braw file directly for the gyro data.

Thanks once again, CaptainHook. I ran tests with zooming 11>20mm, first I set the camera metadata to a focal length of 11mm and then the same 11>20mm zoom with camera metadata set to a focal length of 20mm.

To me the setting focal length to the end of the zoom seemed better, but it might just be because I did a better job on the second zoom.

Thanks a lot for the detailed reply, dear Captain.

i did some small test, and compare different stabilization way.

In this link you can find the result and in the comment you can see the link to download that videos and try yourself.

It's very important to notice that gyro give you less rolling shutter distortion in many exaggerated situation of motion than other stabilization method.
to me, gyro way is a game changer, obviously is not a magic tool, i need to walk stable and not ask to stabilize the impossibile, but do a lots of good work.

thanks for the test Carlo. Also note the less zoom with Gyro... and it likely takes less time to compute. Nice to have in the pocket 4K, but can't wait for it to arrive in the 12K with the other improvements... like RGB histogram. Oh, and please, please make the record button in the viewer not red when not recording... only red when recording... and other color when not recording would help.