PDA

View Full Version : Rolling Shutter Test



markmwilliams
09-09-2012, 05:50 AM
There's been a lot of talk about rolling shutter artefacts on the BMCC but no one seems to have demonstrated how problematic it could be.

Would it be possible for someone with the camera to film some naked and handheld BMCC with vertical lines in b/g on something like a 50mm lens?

I'm not looking for charts or scientific comparisons or even super fast whip pans but a real world test of things we may all be shooting soon.

For me this info would help decide if I switch to the MFT mount - ie: if IS is going to be vital for handheld work on the BMCC.

Maybe someone has already posted about this and I'm not aware of it but otherwise if someone could oblige I (and I'm sure others) would be hugely grateful. :)

Thanks M

nickjbedford
09-09-2012, 05:56 AM
Yeah, I'd like to actually see a test!

It keeps getting mentioned but I hardly see any in any of the videos that are posted, even John Brawley's few videos which have some moderate movement in them.

morgan_moore
09-09-2012, 05:58 AM
Some one maybe PB, cant remember, had a shot of a underground train leaving the station. about 1/4 of a second.

Showed skew.

Overall Ive done loads of carmount/steadicam with my 5d so..

S

mico
09-09-2012, 07:51 AM
No one wants to be the the first to test something that will go viral with their name attached that MIGHT break the magic spell of a camera that hasn't gotten out of the gate yet. Whether its similar to the 5D or better, people will pick it apart to death. Bloggers with test cameras don't want to destroy their relationship with companies with a test video that MIGHT cause a commotion whether that commotion is warranted or not. Its a no win. Personally,John Brawleys short film footage is all I need to know about the camera.

mbeck
09-09-2012, 08:22 AM
I'll shoot some 50mm handheld tomorrow when I get the camera.

Frank Glencairn
09-09-2012, 09:37 AM
Okay, here we go.

First one is on a slider, second one is on a tripod.
24p, ProRes/film, 180 deg shutter, Samyang 35mm

I think it is not as bad as in the DSLRs, and to be honest, in my whole life as a DP, I newer whipped my camera around like that.






http://youtu.be/lx0k98YWxQU

cineman
09-09-2012, 09:54 AM
It is more obvious than I was hoping. Still workable though. Seems better in someways that the dslrs.

Steve4505
09-09-2012, 10:56 AM
It is more obvious than I was hoping. Still workable though. Seems better in someways that the dslrs. I have not worked on just a few lower-end pro cameras, but I thought learning how to work with the limitations of any camera was what you did (not against them). There is nothing I've worked on which would look "good" in those test shots either. Has anyone?

Thanks Frank Glencairn for the test. I agree and I am confused why folk would be moving a camera around fast and actually expect it to "follow" perfectly. There was one part where Frank was panning right relatively slowly across the buildings (I believe on tripod) and I didn't like the look of that though.

It irks me a bit that the off the cuff PB comment about rolling shutter had one Vimeo user say "that's it I am not buying this camera".

Peter J. DeCrescenzo
09-09-2012, 11:02 AM
This may or may not be what Frank has done, but note below what Barry Green requested concerning rolling shutter skew artifact testing. Seems like a pretty straight-forward test. Has anyone done it?



The "wiggle" test is pointless and outdated. There is a very simple easy and truly scientific test possible. Just level the camera and shoot a vertical stripe, panning past it in both directions. Supply two frames back-to-back. From that we can calculate exactly what the rolling shutter is, and you don't need to do side-by-side or multiple cameras simultaneously and you don't need to worry about field of view or panning rate or anything. Sure wish someone would do this, so we could answer this question definitively once and for all.

Jorge De Silva
09-09-2012, 11:04 AM
Better than I expected to be true! That works for me! It's more than ok for a camera under 5k!

mbeck
09-09-2012, 11:11 AM
Yeah, I was hoping for better, but I can't complain! It is absolutely not bad enough to be a major problem.

Cedric Akins
09-09-2012, 11:18 AM
I have not worked on just a few lower-end pro cameras, but I thought learning how to work with the limitations of any camera was what you did (not against them). There is nothing I've worked on which would look "good" in those test shots either. Has anyone?

Thanks Frank Glencairn for the test. I agree and I am confused why folk would be moving a camera around fast and actually expect it to "follow" perfectly. There was one part where Frank was panning right relatively slowly across the buildings (I believe on tripod) and I didn't like the look of that though.

It irks me a bit that the off the cuff PB comment about rolling shutter had one Vimeo user say "that's it I am not buying this camera".

I agree with you for the most part but if you have movement in the shot, NOT JUST THE CAMERA MOVING, can cause your shot to look bad or unusable.

Brian@202020
09-09-2012, 11:18 AM
Wow! To me that looks pretty bad Frank. The 5D2 and GoPro are worse for sure, but I feel most other CMOS broadcast and DSLR camera's are better.

Gwangjuboy
09-09-2012, 11:24 AM
Okay, here we go.

First one is on a slider, second one is on a tripod.
24p, ProRes/film, 180 deg shutter, Samyang 35mm

I think it is not as bad as in the DSLRs, and to be honest, in my whole life as a DP, I newer whipped my camera around like that.











http://youtu.be/lx0k98YWxQU


How is showing the camera on a footage show the reality of jello in what is from the CEO's own words meant as primarily to be handheld. You said recently that you didn't want to show some handheld footage because it basically sucked. The guy above is right people don't want to be the one that shows what could be a serious negative with this camera. Even your on tripod test looks a little worrying actually. It's important for people to know before they choose what mount, an EF with IS may be the only way to be handheld with this camera.

Brian@202020
09-09-2012, 11:29 AM
Maybe they will be able to speed up the refresh rate of the sensor in a firmware update. RED did it with the RED ONE. The RED ONE was really bad at first, and 2 or 3 firmware updates later it was considerably better.

Frank Glencairn
09-09-2012, 11:31 AM
How is showing the camera on a footage show the reality of jello in what is from the CEO's own words meant as primarily to be handheld. You said recently that you didn't want to show some handheld footage because it basically sucked. The guy above is right people don't want to be the one that shows what could be a serious negative with this camera. Even your on tripod test looks a little worrying actually. It's important for people to know before they choose what mount, an EF with IS may be the only way to be handheld with this camera.


I said to before, and I say it again. If you think you can take the bare camera out of a backpack and wave it around, while holding it in your hands (tourist style), you will be disappointed. It's a cinema camera and if you treat it like one, you gonna love it.

It may look like a DSLR, but it needs a monitor, a battery, a field mixer, a tripod or good shoulder rig (and a DP with a very stable hand) and a rig where you mount all that.

Yuns
09-09-2012, 11:31 AM
Frank, would it be possible to share the original DNG or Prores of the scene or to share two frames in a row from the scene. From two back to back frames showing skew we can calculate the sensor read out speed by comparing pixel movement in the vertical line.

At reduser, they've measured the Scarlet at normal resolution at 14 ms. I've seen measurements for the FS100 at 15 ms and AF100 at 14.85 ms. In contrast, the D90 was 33 ms, 5D Mark II was 25 ms, 7D 21 ms, GH1 25 ms and EX1 16 ms according to Provideocoalition. Tessive, the aftermarket shutter company, measured the C300 at 16 ms, the F65 at 14 ms (though the F65 has a mechanical shutter to address rolling shutter issues), and Red One MX at 4K at 16.6 ms (others have measured it as 15 ms). The 5D Mark III and D800 allegedly have half the rolling shutter of the 5D Mark II but I haven't seen any actual measurements yet.

Therefore, I'd expect the BMCC to fall between the 25 ms of the 5D Mark II and the 15 ms of the FS100 based on what John and others have said. So maybe 20 ms give or take but I don't know until we can see still frames.

Cedric Akins
09-09-2012, 11:37 AM
How is showing the camera on a footage show the reality of jello in what is from the CEO's own words meant as primarily to be handheld. You said recently that you didn't want to show some handheld footage because it basically sucked. The guy above is right people don't want to be the one that shows what could be a serious negative with this camera. Even your on tripod test looks a little worrying actually. It's important for people to know before they choose what mount, an EF with IS may be the only way to be handheld with this camera.

Coming from an audio guy, that footage is not that bad. It can be fixed in post via software. But like Frank said it is rare that you would ever whip pan your camera in that manner. IMHO the effect actually adds to the shot. If this was a shot in a production the speed of the shot would suggest dizzying emotion.

Felix
09-09-2012, 11:48 AM
Ok, thatīs my Fs700. It were the only vertical lines I had available without driving somewhere (and since itīs sunday Iīll stay in my underwear ;) )
I think the FS700 is one of the better rolling shutter cameras because of its 240 fps readout.
50mm (should be like 35 mm on the BMCC)

https://vimeo.com/49112537

How much better is this?

Edit: Yuns, can this be used for your calculations? I can upload the mts file

Jorge De Silva
09-09-2012, 11:52 AM
Ok, thatīs my Fs700. It were the only vertical lines I had available without driving somewhere (and since itīs sunday Iīll stay in my underwear ;) )
I think the FS700 is one of the better rolling shutter cameras because of its 240 fps readout.
50mm (should be like 35 mm on the BMCC)

https://vimeo.com/49112537


How much better is this?

Edit: Yuns, can this be used for your calculations? I can upload the mts file

It is me... or it is even worse?!

Cedric Akins
09-09-2012, 11:52 AM
Looks to me that the BMD CC is marginally worse, IMHO. But not by much. If you look at price then the FS700 is worse. For what it costs it should perform better.

dop16mm
09-09-2012, 11:57 AM
the vertical leaning in whip pans, doesn't bother me. Jello wobble does, footage released so far has been careful not to show that. My biggest frustration with CMOS cameras is not being able to just pick it up and grab a shot, and I'm not talking Bourne action, just not tripod. I can hold my breath and hope for the best, and sometimes it will be gold and sometimes the slightest twitch will ruin the shot. Maybe the actual weight of this will help some.

Yuns
09-09-2012, 11:57 AM
Ok, thatīs my Fs700. It were the only vertical lines I had available without driving somewhere (and since itīs sunday Iīll stay in my underwear ;) )
I think the FS700 is one of the better rolling shutter cameras because of its 240 fps readout.
50mm (should be like 35 mm on the BMCC)

https://vimeo.com/49112537

How much better is this?

Edit: Yuns, can this be used for your calculations? I can upload the mts fileSure it could be used. All I would even need is two frame captured in a row at the point of maximum skew in one direction and two frames in a row in the opposite direction. This may help you not have to upload the entire mts. Also which framerate was the video taken at?

I'll try to do a calculation using the MP4 from vimeo first.

Felix
09-09-2012, 12:13 PM
uploading .MTS now.
It was shot at 23,976 fps, 1/50 shutter

But most important is the original prores file from Frank!

Yuns
09-09-2012, 12:26 PM
EDIT: I recalculated see correction below.

Felix
09-09-2012, 12:28 PM
Original MTS (http://www.steinhardtverlag.de/Felix/rs.MTS)

Cornelius
09-09-2012, 12:29 PM
Okay, here we go.

First one is on a slider, second one is on a tripod.
24p, ProRes/film, 180 deg shutter, Samyang 35mm

I think it is not as bad as in the DSLRs, and to be honest, in my whole life as a DP, I newer whipped my camera around like that.






http://youtu.be/lx0k98YWxQU

Agreed, this is certainly less than my DSLR.

Yuns
09-09-2012, 12:36 PM
Recalculated. I may have accidentally compared the wrong lines. This time I'm getting (69 skew/224 horizontal movement)/23.976 for 13 ms read time for the FS700 which is in line with the FS100.

Felix
09-09-2012, 12:40 PM
That should be right. I just tried 100mm hand held without IS. No problems at all unless you really shake it.

mico
09-09-2012, 01:07 PM
Ok, thatīs my Fs700. It were the only vertical lines I had available without driving somewhere (and since itīs sunday Iīll stay in my underwear ;) )
I think the FS700 is one of the better rolling shutter cameras because of its 240 fps readout.
50mm (should be like 35 mm on the BMCC)

https://vimeo.com/49112537

How much better is this?

Edit: Yuns, can this be used for your calculations? I can upload the mts file

Much better

Frank Glencairn
09-09-2012, 01:39 PM
Frank, would it be possible to share the original DNG or Prores of the scene or to share two frames in a row from the scene. From two back to back frames showing skew we can calculate the sensor read out speed by comparing pixel movement in the vertical line.

Can't upload the whole files, they are too big, but here are the frames: http://www.filedropper.com/frames_1

Yuns
09-09-2012, 01:49 PM
Thanks so much Frank. I'll have the analysis done in a bit.

Cornelius
09-09-2012, 01:58 PM
This was me to my horror when I checked the footage/frames. - http://i.imgur.com/4l4lZ.gif

Not really, it was pretty average. Nothing for a guy like me to worry about.

Felix
09-09-2012, 02:00 PM
This was me to my horror when I checked the footage/frames. - http://i.imgur.com/4l4lZ.gif


As a child, I was soooo scared of Nagilum^^


BTW: I canīt stand it Yuns! Give the result! ;)

markmwilliams
09-09-2012, 02:03 PM
Hi Frank,

Thanks for posting the test, really apprecated and it's pretty much what I was expecting.

Yes, this camera has rolling shutter but from the looks of it, nothing that can't be worked around.

The one outstanding question is Jello from handheld. I'm not talking tourist cam - but hand held is a valid cinematic style and it would be good to know that the BMCC holds up well, even using something like a shoulder rig.

Colour me reassured from the dolly and tripod stand point.

Frank Glencairn
09-09-2012, 02:06 PM
Okay, since you guys twisting my arm, I do some shoulder and handheld stuff tomorrow.

Frank

Cornelius
09-09-2012, 02:12 PM
Okay, since you guys twisting my arm, I do some shoulder and handheld stuff tomorrow.

Frank

Thanks Frank, I know some people are very interested to see the results.

Yuns
09-09-2012, 02:15 PM
The first images of the building aren't really usable since the camera is accelerating during the shot (skew is not even shot to shot).

Second image is tough to use since the lines are so blurred but I am getting preliminary results of 25 ms. I would estimate my degree of error as +/- 3 ms. So it could be as short as 22 ms or as long as 28 ms. This is 5D Mark II/7D range. Please don't use my calculation as definitive or scientific and please don't use it as a reason to decide on how bad the rolling shutter is on the camera. My initial calculation on the FS700 was vastly incorrect until I chose different frames to measure. But I wanted to at least determine a point of reference.

Mickael D
09-09-2012, 02:17 PM
Yeah. Looks sad.

http://www.thefoundry.co.uk/products/rollingshutter/

Frank Glencairn
09-09-2012, 02:18 PM
Wait a sec, I get you better frames.

Here you go: http://www.filedropper.com/frames_2

Mickael D
09-09-2012, 02:23 PM
Frank, could you upload some prores footage, generic stuff with some skintones?

Yuns
09-09-2012, 02:26 PM
Thanks Frank! Your effort is much appreciated. To give people an idea of how misleading the calculations can be depending on quality of the frame, if you use the accelerating building shot to run the calculation. You end up with a read speed of 2 ms to 11 ms.

Yuns
09-09-2012, 03:09 PM
I'm officially giving up. I just can't get a consistent read. On the second set of building shots I get a 18.5 ms calculation which seems about right but on the second set of window frames I get about 33 ms which is clearly wrong.

Frank Glencairn
09-09-2012, 03:14 PM
I think it heavily depends on the speed of movement and the way you move the camera.
One shot (building) was a pan on a tripod, the other one (lanterns) on a slider.

But yeah, from my feeling, somewhere between 18 and 20 seems right.

Yuns
09-09-2012, 03:33 PM
Thanks again Frank. Much appreciated.

If anyone else wants to try. The math is simple. Chose a vertical line in the image that is absolutely vertical while still. Move the camera at a consistent speed and choose two frames in succession. Look at the vertical line at the very top of the frame and see the pixel's horizontal location (Frame 1 top pixel X axis location) and then the same vertical line at the very bottom of the frame (Frame 1 bottom pixel X axis location). Now look at the second frame. and do the same for the exact sa,e vertical line. You will get (Frame 2 top pixel X axis location) and (Frame 2 bottom pixel X axis location) .

Skew will be
(Frame 1 top pixel X axis location) minus (Frame 1 bottom pixel X axis location)
(Frame 2 top pixel X axis location) minus (Frame 2 bottom pixel X axis location)
These two should be roughly the same. I averaged them. Let's call this SKEW

Now horizontal panning speed in pixels moved will be
(Frame 1 top pixel X axis location) minus (Frame 2 top pixel X axis location)
(Frame 1 bottom pixel X axis location) minus (Frame 2 bottom pixel X axis location)
These two should also be roughly the same. I averaged them. Let's call this NUMBER OF PIXELS PANNED

So now you have framerate (Shutter speed isn't directly relevant). Each frame begins 1/framerate seconds apart from the next frame.
We also know that in 1/framerate seconds the camera has panned NUMBER OF PIXELS PANNED with the amount of pixels of rolling shutter as SKEW

We know that the horizontal pan speed in pixels per second is NUMBER OF PIXELS PANNED*FRAMERATE
We know its taken SKEW pixels to begin reading the last line of the frame.

So the number of seconds it took to from the first line to the last line of the frame is SKEW/(NUMBER OF PIXELS MOVED*FRAMERATE)
This leaves you with read time in seconds.

Frank Glencairn
09-09-2012, 03:34 PM
Found some clips, that I shot last week in Berlin while our lunch break, just playing with the bare camera handheld and trying out some functions.

That should give you guys an idea, regarding handheld.


http://youtu.be/lQ3eECtnJmM

John Brawley
09-09-2012, 03:53 PM
Hi Frank,

Thanks for posting the test, really apprecated and it's pretty much what I was expecting.

Yes, this camera has rolling shutter but from the looks of it, nothing that can't be worked around.

The one outstanding question is Jello from handheld. I'm not talking tourist cam - but hand held is a valid cinematic style and it would be good to know that the BMCC holds up well, even using something like a shoulder rig.

Colour me reassured from the dolly and tripod stand point.

Like this ?

Mostly HANDHELD (http://johnbrawley.wordpress.com/2012/08/07/some-more-blackmagic-cinema-camera-footage/)

jb

Frank Glencairn
09-09-2012, 03:55 PM
John seems the have a more steady hand than I do :)

nickjbedford
09-09-2012, 03:58 PM
A shoulder rig would help a lot with the micro jitters. Thanks for posting some tests. It's a lot clearer now.

dop16mm
09-09-2012, 04:22 PM
I Believe John's shot is in a rig and looks pro, no issues. Frank's naked hand-held was also quite good, and mostly within acceptable tolerances. The jitters did not seem to induce rolling jello, good to know.

markmwilliams
09-09-2012, 05:10 PM
Frank and John,

Thank you both for posting this footage. Looking at it I agree with Nick that a shoulder rig will be really useful for controlling micro jitters and I now have very few concerns over handling the rolling shutter - certainly not as bad as DSLR (which is pretty impressive considering the crop factor doesn't help matters)

Plus Frank, your quick handheld footage, micro shakes aside, still looked fantastic. I'm really loving the aesthetic of the camera.

And John is either made of solid rock or he never drinks coffee or booze. Never seen hands so steady.

Now roll on December so I can finally get my hands on this camera!!!

Frank Glencairn
09-09-2012, 05:16 PM
Plus Frank, your quick handheld footage, micro shakes aside, still looked fantastic. I'm really loving the aesthetic of the camera.



Yup, it has a certain something to it, that really won me over.
Though I still stand behind of the most points I made in my first BMC article right after NAB (mostly regarding ergonomics), for the picture this camera delivers, I'm happy to work around all that.

Frank

mbeck
09-09-2012, 05:45 PM
Thanks Frank! definitely better than my 7D! Makes me a happy camper!

Steve4505
09-09-2012, 06:31 PM
Phillip Bloom has some the some of his test footage shot in Finland I think using "target shooter" mini-rig (or one like it) from Zacuto. JB and Franks hand held footage is as good or better as the other pro's I've worked with (I am no where near as good). I sure wish A LOT more video ops could be as steady, because I've seen quite a bit of footage used on a variety of projects which was much much worse. I am not very concerned now about the BMC and handheld jello. I would still like to see slow pans on a tripod, because I can't recall any of the tests using them.

939

nihilessence
09-09-2012, 08:37 PM
Just a note for people new to The Foundry's Rolling shutter (or PFTrack) I have used it and while it does work for many things, it depends on optical flow, so there are situations it won't work so well just like Twixtor doesn't work in certain situations - it depends on optical flow too. Also, the greater the accuracy of the RS index the better it works.

It'd be great if manufacturers released the statistic on how quickly each consecutive pixel is activated.

EDIT: The thing about CMOS correction is it was invented for 3-D tracking in VFX - not for the aesthetics of hand held CMOS.

nihilessence
09-09-2012, 08:49 PM
Thanks Frank! Your effort is much appreciated. To give people an idea of how misleading the calculations can be depending on quality of the frame, if you use the accelerating building shot to run the calculation. You end up with a read speed of 2 ms to 11 ms.

Wouldn't you need a constant velocity to get a correct read? If you're accelerating, the camera is moving faster by the time the line refresh has worked its way down to the bottom of the sensor.

Joe Giambrone
09-09-2012, 09:19 PM
That is removable though, no? There have been plugins for years to address that kind of skew due to panning.

Yuns
09-09-2012, 09:33 PM
Wouldn't you need a constant velocity to get a correct read? If you're accelerating, the camera is moving faster by the time the line refresh has worked its way down to the bottom of the sensor.Yes, you need a constant pan velocity within the sets of frames you are reviewing. I was just using that example to show how acceleration and deceleration can throw off the calculation tremendously.

Yuns
09-09-2012, 09:50 PM
As a side note James Tonkin of Hangman Film just posted this video to Vimeo which he shot handheld with no support using only a Canon EF-S 17-55mm lens. Looks pretty decent.

https://vimeo.com/49059708

Steve4505
09-09-2012, 10:34 PM
As a side note James Tonkin of Hangman Film just posted this video to Vimeo which he shot handheld with no support using only a Canon EF-S 17-55mm lens. Looks pretty decent.

https://vimeo.com/49059708 Loved it.

Lorenzo Straight
09-09-2012, 10:53 PM
So, just stay away from 'Blair Witch' style shooting.

Martin
09-10-2012, 12:37 AM
This was shot handheld, posted in a new thread on the forum but thought it belongs in this thread as it shows well the amount of jutter we can expect shooting handheld without any support gear:
https://vimeo.com/49059708#at=0

still, lovely skintones and detail!

Soeren Mueller
09-10-2012, 01:09 AM
This was shot handheld, posted in a new thread on the forum but thought it belongs in this thread as it shows well the amount of jutter we can expect shooting handheld without any support gear:
https://vimeo.com/49059708#at=0

still, lovely skintones and detail!

Nice! That's exactly the 16mm handheld look I was hoping for. Now if somehow the optical stabilizer would be supported/activated on EF lenses that would help a lot.. on the other hand I haven't seen any wide prime or zoom for EF with OIS :( .. but from my DSLR experience I know that it helps a looot to get rid of the "micro judder RS wobble" if you want to go handheld. But perhaps a little shoulder support rig is enough on the BMC to get rid of that.

Although it's completely in another ballpark pricewise it was nice to try out the Ikonoskop cam on IBC as it doesn't have any RS at all. Unfortunately it's sort of a no-brainer when it comes to the overall cost ;)

David
09-10-2012, 02:10 AM
Something to consider in regards to rolling shutter. You have two people running towards each other for that epic embrace in your story at the train staion with one train speeding left and another speeding right. You have other people both in the foreground in front of the trains and in the background behind the trains because this has to look legit. To top it off you are on dolly or crane because you want good production value! Which rolling shutter plugin is going to fix this? You have skews going in 3 separate directions and since the camera is moving along with the trains a 2D fix looks fake. Sometimes depending on the foreground and background elements along with what is moving, fixing rolling shutter goes way beyond what your typical 200 or 300 dollar an hour after effects or nuke artist will be able to fix. Fixing skew issues properly on certain shots goes far beyond creative masking and rotoscoping. Sometimes fixing these types of problems requires matte painters both 2D and 3D, 3D Animators, a Flame/Inferno artist, essentially a post VFX team to fix. Things to consider in how you frame your WOW factor shots!

This in no way shape or form is going to make me cancel my order but it is something that I have become very watchful of having shot on various CMOS cameras that read out progressively. Sorry to be so repetetive but without careful preperation and proper framing & composition along with a choreographed sequence of events, you will have shots that you will either need to accept as they are or have to spend upwards of what your entire shooting budget was to properly fix if you are working on small scale projects like me. This is not to knock The Mill as they are extremely good at what they do, but just to demonstrate the issue, we had a get our epic train shot fixed and the estimate was $75,000. We only had $15,000 budgeted for the entire shoot and $10,000 for post including sound, edit and gaphics! There was no way we could have afforded this. Even if we could have afforded it we would have been pushed back 3 months because rightly so they were working on more important stuff than our low budget production. I can't hold it against them that they have high profile high paying clients. We went to 3 different post houses in NYC and were basically told you need to go to ILM, DIGITAL DOMAIN, WETA or some other big name place to get this fixed to the level that we were looking for. Eventually somebody else took over the reigns of finding a smaller VFX house to fix it and got pretty much similar responses as we got from The Mill. The client ultimatley was not happy about the rolling shutter. He said he didn't have any of those issues on their previous project which was shot on a Sony digi-beta and was surprised because he thought "HD" would be better. We had rented a Red One BTW. I'm not sure if there are any location scouts here but we had to get so many favors pulled, permits, extra insurance, letters to the right people etc to be able to shoot at a train station and still make a profit. You know... friends and family come the shoot because you can't afford to pay all the extras type stuff. We only had the opportunity for 2 takes. A re-shoot with a differently framed shot was entirely out of the question and had to live with the shot. I honestly was not that bothered by the skew as I found it to be minimal but the client asked why are the trains lopsided? We never got a job form them again.

Frank Glencairn
09-10-2012, 02:19 AM
I always wondered why they shot something like "24" mostly on film, and telling me, it's cheaper at the end.
Now I start understanding. Thanks for the heads up and sharing that story David .

Frank

Soeren Mueller
09-10-2012, 02:22 AM
I can only second that!
"We'll fix it in post" just doesn't really work for rolling shutter, only if it's a super simple shot basically.

@Frank.. either film or a CCD based cam or CMOS + global shutter ;)

Frank Glencairn
09-10-2012, 02:32 AM
Oh boy, I would really like to see more cameras with mechanical shutter .
I played with the F65 and really liked it. Looks like the DigiBolex would have some use in certain situations, if the rest of the camera is up to the quality we need.

Martin
09-10-2012, 02:33 AM
Not sure if you were referring to 'me' with 'you', but just to clarify I was just re-posting someone's work. I do not have a BMCC yet...

nihilessence
09-10-2012, 02:43 AM
Something to consider in regards to rolling shutter. You have two people running towards each other for that epic embrace in your story at the train staion with one train speeding left and another speeding right. You have other people both in the foreground in front of the trains and in the background behind the trains because this has to look legit. To top it off you are on dolly or crane because you want good production value! Which rolling shutter plugin is going to fix this? You have skews going in 3 separate directions and since the camera is moving along with the trains a 2D fix looks fake. Sometimes depending on the foreground and background elements along with what is moving, fixing rolling shutter goes way beyond what your typical 200 or 300 dollar an hour after effects or nuke artist will be able to fix. Fixing skew issues properly on certain shots goes far beyond creative masking and rotoscoping. Sometimes fixing these types of problems requires matte painters both 2D and 3D, 3D Animators, a Flame/Inferno artist, essentially a post VFX team to fix. Things to consider in how you frame your WOW factor shots!

This in no way shape or form is going to make me cancel my order but it is something that I have become very watchful of having shot on various CMOS cameras that read out progressively. Sorry to be so repetetive but without careful preperation and proper framing & composition along with a choreographed sequence of events, you will have shots that you will either need to accept as they are or have to spend upwards of what your entire shooting budget was to properly fix if you are working on small scale projects like me. This is not to knock The Mill as they are extremely good at what they do, but just to demonstrate the issue, we had a get our epic train shot fixed and the estimate was $75,000. We only had $15,000 budgeted for the entire shoot and $10,000 for post including sound, edit and gaphics! There was no way we could have afforded this. Even if we could have afforded it we would have been pushed back 3 months because rightly so they were working on more important stuff than our low budget production. I can't hold it against them that they have high profile high paying clients. We went to 3 different post houses in NYC and were basically told you need to go to ILM, DIGITAL DOMAIN, WETA or some other big name place to get this fixed to the level that we were looking for. Eventually somebody else took over the reigns of finding a smaller VFX house to fix it and got pretty much similar responses as we got from The Mill. The client ultimatley was not happy about the rolling shutter. He said he didn't have any of those issues on their previous project which was shot on a Sony digi-beta and was surprised because he thought "HD" would be better. We had rented a Red One BTW. I'm not sure if there are any location scouts here but we had to get so many favors pulled, permits, extra insurance, letters to the right people etc to be able to shoot at a train station and still make a profit. You know... friends and family come the shoot because you can't afford to pay all the extras type stuff. We only had the opportunity for 2 takes. A re-shoot with a differently framed shot was entirely out of the question and had to live with the shot. I honestly was not that bothered by the skew as I found it to be minimal but the client asked why are the trains lopsided? We never got a job form them again.

I haven't seen the shot, but I can tell you it's possible to fix a rolling shutter FUBAR without spending that kind of money. Yes it might take a bit of VFX time, but even if you have to buy the software and do it yourself, you're not going to get anywhere near that amount. I suspect they were trying to fob you off politely.

As with a lot of things in filmmaking it's cheaper to find a solution before anyone says 'action'. But as an example, District 9 was shot on the Red One and The Embassy was tasked with match moving the power suit shots. Just have a look at the hand held movement on that show. It does help that the camera was loaded up with a lot of accessories.

kbmediaonline
09-10-2012, 02:45 AM
The only dslr camera Ive owned is a T2i which Im sure has really bad rolling shutter. Ive shot almost 30 music videos with lots of storylines and even action scenes and never had to worry about my shots looking like jello. Why is this such a big issue? And when will you be moving your camera back and forth that fast and for what reason?

Cornelius
09-10-2012, 02:46 AM
This was shot handheld, posted in a new thread on the forum but thought it belongs in this thread as it shows well the amount of jutter we can expect shooting handheld without any support gear:
https://vimeo.com/49059708#at=0

still, lovely skintones and detail!

Watched this about an hour ago, I was very impressed with the handheld shots. You can tell the weight really helps keep it steady and sometimes if it's just for fun, you only want to get the camera out, nothing else.

David
09-10-2012, 03:04 AM
I haven't seen the shot, but I can tell you it's possible to fix a rolling shutter FUBAR without spending that kind of money. Yes it might take a bit of VFX time, but even if you have to buy the software and do it yourself, you're not going to get anywhere near that amount. I suspect they were trying to fob you off politely.

As with a lot of things in filmmaking it's cheaper to find a solution before anyone says 'action'. But as an example, District 9 was shot on the Red One and The Embassy was tasked with match moving the power suit shots. Just have a look at the hand held movement on that show. It does help that the camera was loaded up with a lot of accessories.

Well being that I spoke with The Mill and 3 different post houses myself along with someone else who also contacted various different post houses and all came with pretty much the same answer, I tend to believe them. Pretty much the whole shot would be deconstructed lots of people would need be rotoscoped about 37 alltogther. The trains would be modeled in 3D because since the train move and the camera move the skewing it back into place just didn't look right. No skewing it back amount looked right. The original trains would painted out, with the new ones put in. and the people behind the trains would be recreated for the frames were the obscuring would be different. This was back in 2008 so I don't know if I still have the estimates but we received price break downs on what would be done itemizing not only what each person was, but also what they would be accomplishing in the shot. It was essentially going to be re-built form the ground up. We looked for cheaper options but either no one wanted what little money we had (willing to sacrifice $5000 of our budget to keep a good relationship with the client) or it was in fact as the multiple different VFX companies had told us.

David
09-10-2012, 03:05 AM
The only dslr camera Ive owned is a T2i which Im sure has really bad rolling shutter. Ive shot almost 30 music videos with lots of storylines and even action scenes and never had to worry about my shots looking like jello. Why is this such a big issue? And when will you be moving your camera back and forth that fast and for what reason?

It's not only moving the camera that causes rolling shutter but the stuff that moves in front of it.

Frank Glencairn
09-10-2012, 03:24 AM
by the way, that can happen even on film (at least back than)

949

David
09-10-2012, 03:27 AM
Thanks for the visual.

Cornelius
09-10-2012, 03:41 AM
by the way, that can happen even on film (at least back than)

949

I love that photo! Lets hope his wheels weren't that shape!

nihilessence
09-10-2012, 04:00 AM
Well being that I spoke with The Mill and 3 different post houses myself along with someone else who also contacted various different post houses and all came with pretty much the same answer, I tend to believe them. Pretty much the whole shot would be deconstructed lots of people would need be rotoscoped about 37 alltogther. The trains would be modeled in 3D because since the train move and the camera move the skewing it back into place just didn't look right. No skewing it back amount looked right. The original trains would painted out, with the new ones put in. and the people behind the trains would be recreated for the frames were the obscuring would be different. This was back in 2008 so I don't know if I still have the estimates but we received price break downs on what would be done itemizing not only what each person was, but also what they would be accomplishing in the shot. It was essentially going to be re-built form the ground up. We looked for cheaper options but either no one wanted what little money we had (willing to sacrifice $5000 of our budget to keep a good relationship with the client) or it was in fact as the multiple different VFX companies had told us.

2008 didn't have The Foundry's or Pixel Farm's rolling shutter solutions, so that explains it. Now, with 3-D tracking, masking and plugins, a lot more can be done on the cheap(ish). Soon Lidar will come in and make it a fully solved problem, IMO.

nihilessence
09-10-2012, 04:03 AM
by the way, that can happen even on film (at least back than)

949

well, the same effect happens to a lesser degree with a mechanical shutter - the image must be revealed bit by bit over time as the edge of the shutter moves out of the way, therefor light hitting the top of the image is farther back in time than that which hits the bottom of the image. It's just that mechanical shutters are much faster (freaking fast).

David
09-10-2012, 04:09 AM
I'm sure most everyone here understands this as it is a simple concept but just in case. Sure you can skew some thing back into alignment but this does not tell the whole story. If you have digital stop watch that refreshed a fast enough, the time being recorded at the top, middle, and bottom of the frame would be different. You are literally at different point in time between the top of the frame and the bottom of the frame. There would be no need to skew anything since no amount of skewing side to side or top to bottom is going to fix that. In one frame you have an image that is transforming over time. What started of on the stop digital watch as a .0001 at the top of the the frame has progressively turned into .0013 at the bottom of the frame. What plug in fixes time changing with in one frame? Some shots are either accepted as they are or recreated the at little to great expense as there is no plugin that skews back time progressively with in one frame, only skews back the image that is left over by rolling shutter which are 2 different things.

nihilessence
09-10-2012, 04:17 AM
I'm sure most everyone here understands this as it is a simple concept but just in case. Sure you can skew some thing back into alignment but this does not tell the whole story. If you have digital stop watch that refreshed a fast enough, the time being recorded at the top, middle, and bottom of the frame would be different. You are literally at different point in time between the top of the frame and the bottom of the frame. There would be no need to skew anything since no amount of skewing side to side or top to bottom is going to fix that. In one frame you have an image that is transforming over time. What started of on the stop digital watch as a .0001 at the top of the the frame has progressively turned into .0013 at the bottom of the frame. What plug in fixes time changing with in one frame? Some shots are either accepted as they are or recreated the at little to great expense as there is no plugin that skews back time progressively with in one frame, only skews back the image that is left over by rolling shutter which are 2 different things.

if the software has a 3-D model of the scene (i.e. all surfaces available to the camera) and how they are moving relative to the camera, and how the camera is moving, and the rolling shutter rate, using a combination of all that with optical flow and a bit of user input would make almost all RS issues trivial to solve. The irony is that when that code is written the CMOS cameras will all probably be global shutter like f65.

In the mean time, it's amazing what you can do with the foundry's rolling shutter software: http://www.youtube.com/watch?v=Zt0u9hsPuZY

nickjbedford
09-10-2012, 04:24 AM
Loved it.

Wonderful! This is the lens I'll be using for a while so it's great to see how it fares.

funwithstuff
09-10-2012, 04:52 AM
This was shot handheld, posted in a new thread on the forum but thought it belongs in this thread as it shows well the amount of jutter we can expect shooting handheld without any support gear:
https://vimeo.com/49059708#at=0

still, lovely skintones and detail!

Great video. Just FYI, you can spot evidence of stabilisation in a few shots here, in the "locked off" shots such as the shot of the policeman's helmet. Could be native FCP X or Premiere's Warp Stabiliser, or maybe something like CoreMelt's Lock & Load. Most shots don't look stabilised, though, and there seems to be no evidence of jelly wobbling.

(Full disclosure: I've made a number of promo videos for CoreMelt on Creative Cow, so I'm hardly impartial, but Lock & Load is really fast and works well.)

nickjbedford
09-10-2012, 05:18 AM
I think you really just need to be aware of the rolling shutter so that you can plan your shots and action to avoid it. There's so many ways to film the same thing (it's an art, not a science).

If the rolling shutter is going to be a definite problem for something intense, then it's time to either rework what you need to shoot or work out a way of fitting in some RED or C300 hire (or ARRI).

But for most of what we'll do with it, I think it'll be a very minor issue in the hands of a good DP (and director).

nihilessence
09-10-2012, 05:33 AM
Good 'body mounted' camera movement is as much an art as the playing of a musical instrument. When it's done to the highest degree it gives the scene a really orchestrated feel even though the camera operator was reacting to the actors - actually reminds me of JB recently talking about how the camera is another character in the scene.

I think it has to do with the principal of 'anticipation' in animation, so the camera operator is reading where the actor is preparing to move next.

So smooth movement is really down to the experience-level of the camera operator (and it helps to put some weight on the rig).

nickjbedford
09-10-2012, 05:50 AM
Good 'body mounted' camera movement is as much an art as the playing of a musical instrument. When it's done to the highest degree it gives the scene a really orchestrated feel even though the camera operator was reacting to the actors - actually reminds me of JB recently talking about how the camera is another character in the scene.

I think it has to do with the principal of 'anticipation' in animation, so the camera operator is reading where the actor is preparing to move next.

So smooth movement is really down to the experience-level of the camera operator (and it helps to put some weight on the rig).

I did a bit of shoulder cam in the music video I'm working on. It's a really nice way of adding subtle reactionary movement into a shot. I really like how it's used in a lot of films.

nihilessence
09-10-2012, 06:03 AM
I did a bit of shoulder cam in the music video I'm working on. It's a really nice way of adding subtle reactionary movement into a shot. I really like how it's used in a lot of films.

Some of the best is in Kubrick films IMO. Kubrick often did the handheld himself and he's a stickler for composition so he'd move the camera to the head of the actor. Funny thing is you don't notice this obsessive movement unless you're watching out for it. That has a lot to do with the skill of the acting too, I guess. Another example of incredible hand held is Tree of Life. I don't think they used any steadicam in that according to the American Cinematographer article. Malick is all about those serendipitous movement improvisations.

Soeren Mueller
09-10-2012, 07:51 AM
Another example of incredible hand held is Tree of Life. I don't think they used any steadicam in that according to the American Cinematographer article. Malick is all about those serendipitous movement improvisations.

While I agree with all of the above, I have to correct you on that one ;) .. Jörg Widmer was the steadicam op on Tree of Life, so there was definately some use of steadicam:

http://www.imdb.com/media/rm2861283584/nm0002152

http://www.imdb.com/media/rm3062610176/nm0002152

nihilessence
09-10-2012, 09:27 AM
While I agree with all of the above, I have to correct you on that one ;) .. Jörg Widmer was the steadicam op on Tree of Life, so there was definately some use of steadicam:

http://www.imdb.com/media/rm2861283584/nm0002152

http://www.imdb.com/media/rm3062610176/nm0002152

Ah right, I was thinking of Pfister on Inception and TDK. No stradicam just dolly. Thanks for the correction.

rizibo
09-10-2012, 01:27 PM
If the it is true that the bmcc has a global shutter then maybe a firmware update can make a global shutter mode to get rid of rolling shutter in fast camera movements. This may decrease the the dynamic range. It may not be possible because of higher data processing. Having global shutter more and 60p burst mode slow motion would make this a perfect camera for me.


http://www.eoshd.com/content/7969/blackmagic-cinema-cameras-sensor-has-global-shutter-made-by-bae-systems

Barry Green
09-10-2012, 02:23 PM
well, the same effect happens to a lesser degree with a mechanical shutter - the image must be revealed bit by bit over time as the edge of the shutter moves out of the way, therefor light hitting the top of the image is farther back in time than that which hits the bottom of the image. It's just that mechanical shutters are much faster (freaking fast).
That car picture is misleading in the extreme, because it's from a still camera, not a movie camera, and it's from a still camera with an extremely narrow shutter.

You would never, EVER get that type of skew on a movie camera, unless maybe you where shooting with a 2-degree shutter or something ridiculous.

The Single-Chip Camera Evaluation (SCCE) camera test from the ASC put that "film is like a rolling shutter" myth to rest. Film absolutely does not skew or lean like a rolling shutter CMOS camera. They performed extensive tests using a rotating target, and the film was almost entirely rock-solid, as it should be, and so was the Alexa. All the other rolling-shutter cameras showed rolling shutter distortion that was absolutely just not there in the film or CCD cameras.

Frank Glencairn
09-10-2012, 04:30 PM
That car picture is misleading in the extreme, because it's from a still camera, not a movie camera, and it's from a still camera with an extremely narrow shutter.

I posted that with tongue in cheek. Maybe we need a smiley for that :)