cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

R3/R5/R6/R7 IBIS RUINS footage shot at wide focal lengths, I would love software changes, Canon!

EngineerKyle
Contributor

After a year of using the Canon R5 as my primary video workhorse, I am not one to ignore the suggestion that this camera has many idiosyncrasies. I originally switched to this camera due to its excellent autofocus and Raw video capabilities. I wish I could say that I also love the IBIS because there are occasions when it is exceptional. For video, however, it can be more of a curse than a blessing, and I happen to have discovered some workarounds that I would love to share with the community of Canon shooters and perhaps Canon themselves.

Having experience as a technical operations programmer, I am aware that implementing software-level changes for hardware is often a highly complex process. As users, we must consider that when desiring changes to our gear. In many ways, I am more of a technician/engineer than a filmmaker, and I'm sure Canon is tired of hearing criticism from videographers who don't know how the technology even functions under the hood. 

With that out of the way, I have developed some ideas that enable users to "trick" the ibis system into working more effectively for video mode. Upon evaluating this fix, Canon may have some insight into a potential software-level fix that would not require a complete re-engineering of the IBIS mechanism. 

Anyways, here's what over a year of attempting to repair the IBIS for my own use cases has shown: 

The correction of pitch and yaw shaking via the IBIS mechanism is where things begin to fall apart. Even when using IBIS only, the moment you shoot wider than ~26mm, there is rampant corner wobble. The only thing I can think is that the IBIS is, in a way, "too strong," and the wider you go, the more you have to adjust the corners to stabilize the image's perspective. While the corners are indeed stabilized in these wobbly shots, the outcome is distracting and undesirable - it ruins the footage in a manner that cannot be recovered in post-production. Because the lens IS alone does not have any wobble issue on cameras without IBIS, it's safe to assume that even in the coordinated IS modes, the IBIS mechanism is causing the wobble, but it seems to be exacerbated by the lenses corrections. Although it is still present when using wide non-stabilized lenses, the "coordinated control" lens IS seems to make wobble occur at even tighter focal lengths (I had it ruin a shot at 40mm on my 24-105 F4 L RF). 

Put simply, when the IBIS attempts to correct pitch and yaw, it is moving the corners of the sensor forward or back to compensate for shakes. The easiest way to visualize it is if you were to take a photo frame and move one corner further away while moving the opposite corner closer to you: the outcome is called perspective distortion, and the problem worsens the closer you are to the photo frame. Similarly, the perspective artifacts caused by the IBIS become worse as you increase your field of view. 

To summarize, the IBIS wobble is an artifact of perspective distortion, which is the outcome of attempting to correct the Pitch and Yaw camera shake by moving the sensor about those rotation points. Whether using RF stabilized lenses or the IBIS mechanism alone with a non-stabilized lens, this problem will always present itself at wider focal lengths. 

According to Canon's own infographic, coordinated control enabled RF lenses perform correction of pitch and yaw shake with IBIS and lens IS simultaneously. With non-stabilized lenses, the IBIS alone makes those pitch and yaw corrections. With adapted canon ef lenses, however, the camera turns off IBIS corrections on the pitch and yaw axis...and, what do you know: PROBLEM SOLVED! 

I've tested this to be true: my Tamron 24-70 f2.8 VC G2 manages to prevent all IBIS wobble! It has its own set of problems, but corner wobble is not one of them on the R5! The same is true of canons ef-s 10-18mm IS...an ultrawide $250 gem that I use on my R5 all the time! They definitely don’t communicate the gyro data as quickly with the camera as the RF mount alternatives, so fast movements can seem a bit “jerky, but I’d take that over corner wobble any day of the week!

So we've identified our culprit: the IBIS unit correcting for Pitch and Yaw shake is where all the IBIS wobble problems occur! Thusly, modes that disable IBIS pitch and yaw corrections fix the wobble entirely. 

Expanding on this logic, I've put together my own version of Canon's infographic, explaining our circumstances. I’ve only included Pitch and Yaw as the axis because that is truly all that matters regarding the presence of wobble:

 

EngineerKyle_0-1658350994103.jpeg

Canon’s full explanation: 

EngineerKyle_1-1658350995223.png

How to fool your camera into fixing the wobble:

Wide-angle shots do not require as much stabilization as those shot with tighter focal lengths, so the easiest way I have found to fix it in-camera is to use a non-electronic lens and tell the IBIS system that the focal length is wider than it actually is. I have found that the most consistently effective multiplier for this is to widen the camera's perceived focal length by a factor of around ~0.75x, so a 24mm lens uses the IBIS tuning for 16mm and so on. For lenses with full-time mechanical focus mechanisms, you can use a dummy canon ef-rf adapter and focus manually. I have done this on occasion with the sigma Art 20mm f1.4, and it worked exceptionally well when the camera was correcting for a 15mm lens. This all works right now; current firmware, no changes. The problem with this method is that it pretty much rules out all modern focus by wire lenses unless we could override the lens's EXIF data in firmware to tell the camera that a 24mm is actually a 16mm. Perhaps the camera could do this effortless conversion on its own in video mode only, where EXIF data is less imperative than in stills? Stills could remain exactly as they are. 

This solution has been very effective and usable for professional work. I can rely on the ibis to create a relatively smooth image, even handheld, when shooting on the Sumire Primes at wider focal lengths (used with the 20mm T1.5 even while carefully walking with no gimbal)! However, I would LOVE to see this applied in the camera software so I can use my RF autofocus lenses without the stabilization system ruining shots. I can get by as-is using a dummy adapter on my wider sigma primes, but I currently have to choose between autofocus or stabilization; both would obviously be nice!

Expanding this method, if I am shooting at VERY wide focal lengths and wobble continues to occur, I will tell the IBIS mechanism to correct for a 1mm - 4mm lens, and it does a decent job of subtly smoothing micro shakes while not ruining the 12-14mm image. This is a solid solution for older wide-angle lenses with bad barrel distortion. The stabilization effect is less significant, but I find it still subtly compensates for micro shakes. 

Can the RF Coordinated Control Wobbles Be Fixed in Firmware?

Disclaimer: This is about to get very nerdy and theoretical…

Short answer: theoretically, yes, with some technical trickery, but the answer is very complex. According to Canon's marketing materials: the lens corrects pitch and yaw by shifting the X and Y axis of an element that is a short distance from the sensor. Even though the element is not shifting its pitch and yaw, the horizontal and vertical shift causes the light to hit the sensor at a different angle. Interestingly enough, because all of the elements are still facing forward, this does not cause any sort of distortion. Instead, it basically just moves the view around the image circle like a tilt-shift lens. However, correcting the same pitch and yaw movement using the sensor causes the image to exhibit perspective distortion. Anytime the corrective lens elements deviate from being exactly aligned in the pitch and yaw axis to the sensor, the perspective distortion would, in theory, be a problem for video shooting. The only way to correct for pitch and yaw without any risk of perspective wobble is by correcting it with an aligned element in the lens moving vertically and horizontally. Adding pitch and yaw rotation to the sensor causes this distortion every time. At least when the corrections are this significant. Sony cameras do not allow the IBIS mechanism to do ANY corrections in the Pitch and Yaw axis when they are working in tandem with an OSS-enabled lens. I would theorize that this is why. Without OSS lens correction, sony IBIS is barely working when using very wide-angle lenses. Again, I think they saw the technical limitations and simply made the IBIS significantly restrained on the Pitch and Yaw rotation axis. Technically, the Canon IBIS is correcting significantly more, wobble is just an unwelcome artifact that comes with the stabilizing strength.

Fixing this without turning off any axis would require a great deal of technical mastery, but if properly accomplished, it would possibly create the greatest video stabilization system of all time. The easiest implementation would have to be digitally in the camera. Or even in post-production if the gyro data was saved to the video files. Mechanically, it is probably impossible for Canon to correct the wobble while compensating for pitch and yaw rotation with the IBIS mechanism. The sensor would have to move too much to catch the light at the appropriate angle. Thusly, the only way to fix it in its current state is through a digital compensation algorithm that analyzes the gyroscopic data and reprojects the image plane in 3d space....yes, it would be as complicated as it sounds. But it IS possible!

Correcting the IBIS wobble digitally in post would potentially prevent that nasty smearing artifact that is caused by low shutter speed electronically stabilized footage (Looking at you, R5C). This is so because the mechanical elements would still be correcting the shake while the shutter is opening and closing, and the algorithm would simply remove this distortion that the drastic corrections are causing. As I said previously, the wobbly corners are actually stabilized quite well, that’s why the photo modes can produce sharp photos with unprecedently long exposures. The correction of these shakes just causes distracting distortions in video mode. 

Call me a conspiracy theorist, but I'm 90% sure that this is exactly what Olympus/OM Systems is/was doing in their IS mode 2 (or at least on the EM1-III and EM1-X). Their “IS mode 1” has horrible IBIS Wobble, yet IS mode 2 has a slight crop but does not have any IBIS wobble and retains the incredible correction capabilities on the pitch and yaw axis...Literally like a gimbal. 

So that's the really complicated, expensive, and probably R&D-heavy, solution that Canon could implement, or even a third-party developer could implement if we had enough gyro data embedded in the video file. But there are definitely more accessible methods to fix the wobble that are far more feasible for Canon:

The easiest solution for the RF optically stabilized lenses is to allow users to turn off the IBIS and keep the lens IS enabled, which the community has constantly requested. The IBIS is not all bad, though. Knowing that pitch and yaw IBIS corrections are the ONLY cause of wobble, it wouldn't even be necessary to disable IBIS entirely. The best of both worlds comes with turning off Pitch and Yaw axis corrections on the IBIS mechanism while retaining IBIS Roll and X/Y corrections. This is exactly what the camera does for adapted EF stabilized lenses, so why not for RF as well? 

Conclusion

Canon simply needs to treat RF lenses like the compatible EF lenses in video mode: No more Pitch or Yaw corrections from the IBIS mechanism. Let the lens do its good work! 

I would be happy to have a Canon engineer tell me why that’s not possible....which may be the case...but considering it’s already the go-to method for EF lenses, which are older and slower in every way compared to RF, I’d find that hard to believe!

All in all, I am immensely appreciative to Canon for creating such a capable camera! Customers threw a lot of hate it's way, and a great deal of it was misplaced, overlooking that it's a complete hybrid camera with internal raw video and exceptional autofocus. I still can't help but have faith in the company to implement some form of the solutions mentioned in this post. I've outlined all of my ideas here, and while I don't have access to the software, I feel there has to be a way to make it happen, even if outright lying to the camera via EXIF data is the route that would have to be taken, I’m here for it!

6 REPLIES 6

Tronhard
Elite
Elite

Canon has a suggestions site that this will be more likely to find traction on. 

 Actually here is some information from one of my colleagues:
Feature requests should be submitted using the +Feedback option at the bottom the support portal
The link to that site is HERE 


cheers, TREVOR

"All the variety, all the charm, all the beauty of life is made up of light and shadow", Leo Tolstoy;
"Skill in photography is acquired by practice and not by purchase" Percy W. Harris

Hello Tronhard! 

I have tried that, and I have not received a response. Also, it is limited to 1000 words. I am posting this here mainly to help other canon users who are looking for a potential solution to the video wobble issue.

I have sent canon a link to this article so that I can include the full version.

Good luck with that!  Maybe I should do something similar for my feelings about the shortcomings of the R7...


cheers, TREVOR

"All the variety, all the charm, all the beauty of life is made up of light and shadow", Leo Tolstoy;
"Skill in photography is acquired by practice and not by purchase" Percy W. Harris

Thanks for going to all of this effort.

Possibly related, or maybe in addition to what you have noticed, I have noticed that there is a difference between how the EOS R5 does IBIS after power on and after wake from sleep.

If it were code that I had written, I would begin by looking at what might have changed while the camera was sleeping. For example, a zoom lens might have changed focal length. A lens or extender might have powered down and need to interrogated again after it powers up. Some data structures could possibly be reinitialized incorrectly on wake but correctly on power on. The timing of communication between camera and lens might be different during power on than during wake.

Since I only do still photography and seldom use wide angle and use only EF and an adapted Minolta manual lens, it works for me to power the camera off and back on again if it has gone to sleep and I might have changed the focal length of a zoom.

Thanks again.

John

---
https://www.rsok.com/~jrm/

Tronhard
Elite
Elite

I would recommend you post to the VIDEO SOLUTIONS conversations where folks specializing in videos will get the benefit of all you hard work and effort.


cheers, TREVOR

"All the variety, all the charm, all the beauty of life is made up of light and shadow", Leo Tolstoy;
"Skill in photography is acquired by practice and not by purchase" Percy W. Harris

EngineerKyle
Contributor

Great news! Canon just released the in-camera wobble fix algorithm, and I'm happy to report that it does a pretty good job! The fix that I have mentioned above is still necessary for shooting full frame with no crop, or any raw modes. But my RF wide angles are usable for video again and I'm very happy about that. I'm not sure if it's just a coincidence that they released this update after my sending this to them a couple of weeks ago, but I would still appreciate an option to disable pitch and yaw corrections on RF so I can shoot 8k raw again! 

Announcements