Lens breathing problems on 2.5D calculations when pulling focus ... upgrade lens ?

[PatrickFaith]

Background: On 2.5D z-depth calculations, I'm creating 3D meshes off one camera by having a single camera moving on tracks (i.e. the 3d calculation is done between frames as the camera moves position) - this is working really well with the new blender camera tracking software [blender 2.62 ] .

Problem: sometimes I need to change focus while the camera is moving, and this slightly is changing the lens mm, needless to say the 3d interpolation doesn't like the focal length changing while it is trying to calculate depth from camera movement( there really isn't a math correction I know of for mm focal length changes during 2.5d calculations).

So I'm currently shooting with a single scarlet and Zeiss Compact primes, I was thinking i need to upgrade to full cinema glass on shots where i need to have the camera move while pulling focus(i.e. doing the 2.5D z depth calculation) so I don't have any lens breathing.

Question: What is the ... hate to use the word ... but Cheapest glass that has hardly any lens breathing(and the glass must be optically very nice for the calculations). My guess is the RED Prime glass, but I can't figure out how many mm the focal length changes during focus?

Example Test: If I was using a 35 mm RED PRime, and was focusing on something 3 feet away from me (let's say it's actual fcl is 35.002) , and then i focused on 30 feet away ... what would be the delta focal length? Is Red prime real close to .0001 fcl change on focusing, what's the actual number, is their a breathing chart by lens?

Anyway, my assumption is the big movie guys are using high end cinema lenses, and and they a spending big bucks to convert old 2D to 2.5D, but does anyone have numbers for 2.5D breathing errors by lenses? My only other option is complete the 2.5D calculation, hold the camera still, then change focus (i.e. this would mean I couldn't pull focus in the entire movie I'm doing, since about 70% of the movie is generated via the zdepth meshes).

Too icky to think about?

[update ... got some info on the red 17-50 breathing ... it's perty significant ... in the percentage points so NO pulling focus in 2.5d with that. Still don't have numbers for a normal red prime fixed]

 

[Ketch Rossi]

Patrick,

3D, is just about the MOST COMPLEX image capturing way there is.

Any artifacts which are not controlled, and or are really different from lens to lens in the same pair, will cause problems, some more then others,
some can be taken care of in post, some can't.

Some wil cost lots of time and money, some less.

You said it yourself first, the Zeiss compact should not even carry the Zeiss name.

The reason why we used expensive glass for 3D is simple, containing the lens artifacts as much as possible, eliminate Internal Focus shifts,
Breathing, and insuring Even illumination thru Focus range at all apertures and thru out the entire range of the Zoom,
when Zooms are used, which the best are the Rouge 16-42's.

Ultra Primes, are great lenses even so it does take few sets to get the perfect matched set.

Master Primes are way better as their Glass difference is extremely low.

Leica Summilux cine Primes, are absolutely Insane, and carry no artifacts, and almost either lens you pair is a perfect match with minor calibration.

In 3D, both lenses have to be MATCHED, or there is issues to be encountered in post, but post for corrections, need to apply no matter what,
just depends on the glass used and pairing made before end to see how much.

Of course the lens is not all, there is the Mirror, and a variety of other factors, but the lenses are at the very core of it.

 

[Noah Yuan-Vogel]

@Ketch, for the stated purpose, "2.5D" or interpolated 3D, matched lenses don't seem like they would be an issue since he is only using one lens and interpolating the second eye

@Patrick, would it be possible to shoot a little wider, and then pre-process your 2D footage using a software zoom stabilizer or other method that would dynamically crop and compensate for the changes in FocalLength/FOV due to breathing? This seems like it might be more effective than having to find perfect lenses without breathing since I imagine all lenses breathe to some degree, and extensive testing would be required to figure out if there is one lens or select lenses in a set fit the tolerances required for your purpose. I wouldn't be that surprised if you had to do some Focal Length compensation even for the most expensive lenses when changing focus significantly.

I wonder if you also have to be careful about shifts in the lens' entrance pupil during focus changes? Or maybe that is not significant enough to make a difference. That would be much harder to compensate for I imagine.

 

[Ketch Rossi]

@Ketch, for the stated purpose, "2.5D" or interpolated 3D, matched lenses don't seem like they would be an issue since he is only using one lens and interpolating the second eye

@Patrick, would it be possible to shoot a little wider, and then pre-process your 2D footage using a software zoom stabilizer or other method that would dynamically crop and compensate for the changes in FocalLength/FOV due to breathing? This seems like it might be more effective than having to find perfect lenses without breathing since I imagine all lenses breathe to some degree, and extensive testing would be required to figure out if there is one lens or select lenses in a set fit the tolerances required for your purpose. I wouldn't be that surprised if you had to do some Focal Length compensation even for the most expensive lenses when changing focus significantly.

I wonder if you also have to be careful about shifts in the lens' entrance pupil during focus changes? Or maybe that is not significant enough to make a difference. That would be much harder to compensate for I imagine.

The issues are repeated just the same, I stated what 3D doesn't like in a lens, so if using one or two, the problem is different but yet the same,
in the sense that if using two unmatched lenses then the problems will be of different characteristics per each Eye,
if using the same lens, then it is the same issue repeated, but still a problem just the same.

 

[Pedro Guimaraes]

my answer is simple.....rent and test.

 

[PatrickFaith]

@Ketch .... Master Primes are way better as their Glass difference is extremely low.

I think that is the real answer, only the master primes seem to have almost zero breathing durring focus pull.

@Noah ... would it be possible to shoot a little wider, and then pre-process your 2D footage using a software zoom stabilizer or other method that would dynamically crop and compensate for the changes in FocalLength/FOV due to breathing?

I think your right/close doing it the "low end" way, basically if i run the robot twice, once for solving for 3d, second for pulling focus, then merge ... the math will work out. It just takes more time, and the repeat on the robotic shoot has to be perty close.

@Pedro my answer is simple.....rent and test.

Main problem for us opensource/indi math guys, is it takes us months/years of testing/coding ... where we need to run the test, change the code ... then retest (is more a kin to pre-production testing) ... and my problem is i can't affort renting master primes for months on end in pre-production.

Thanks much for your guys help, when I have a solution all post the test.

 

[Bruce Allen]

Patrick, why not rent a ton of SLR lenses from somewhere cheap like lensrentals.com and find out which ones breathe least?

Also:
- what f-stop are you at?
- is the subject moving?

Bruce Allen
www.boacinema.com

 

[PatrickFaith]

Patrick, why not rent a ton of SLR lenses from somewhere cheap like lensrentals.com and find out which ones breathe least?
Also:
- what f-stop are you at?
- is the subject moving?
www.boacinema.com

Hi Bruce, this is for a non-profit project I'm doing (about malnutrition), so they have no insurance during pre-production (100% agree on testing with rental houses for a normal project). As you know, it's real hard to rent without the production insurance. As a side note, the scarlet opens up a lot of opportunities for non-profit low budgets, but the glass side really hasn't changed much and pro-glass really breaks this low end business model. I BTW need to keep this project separate from my other stuff (how the industry works, network stuff is often 2 companies removed from movie production activities anyways).

On the other question:
t-stop is normally around 3, and there is only one subject moving (in about 50% of the shots). Almost all camera moves are done by stepper motors (90% of the shots can be repeated almost exactly with stepper motors), so it's possible to do close to exact re-takes minus the actor. (I was btw pondering testing some Cooke Panchro's since their speed is workable for this situation, but the weight is perty high for the current stepper motor config, Panchro's wouldn't be a master quality ... and it's one of those things where one "upgrade" forces a series of upgrades, etc ...).

 

[Noah Yuan-Vogel]

I think your right/close doing it the "low end" way, basically if i run the robot twice, once for solving for 3d, second for pulling focus, then merge ... the math will work out. It just takes more time, and the repeat on the robotic shoot has to be perty close.

Why would you need to do two passes?

 

[PatrickFaith]

Why would you need to do two passes?

First pass to do the entire 2.5D calculation ( that would give me all the room angles and camera movements) - I'd probably shoot with lights up a bit, and with a high t-stop.
Second pass to do the actual shoot of the actor with the rack focus, and not do any 2.5 calculation.
This way always works, and is standard math.

A one pass solution like u suggested. Could work like this though .... just do a z-map by doing the entire focus range (this also give fcl mm value range change) ... in one or two seconds, which would give me a depth map but not the actual value. Then deal with the error of the breathing knowing i had a general depth solution connected to the mm range. This would be a one pass solution like u suggested. Little bit more math, but would be a faster shoot. The "focus" map btw would be super clean if I had a high end lens like a arri-zeiss master prime. People have been doing focus maps for a long time, but I don't know of a standard open source package that does that? I have a ethernet connector to my remote focus now, so would be fairly easy math, but I don't know what the application protocol is for the remote focus.