Stereoscopic with two 7D's
 

Ok, I stated in another thread that I was going to shoot 3D with the new Pana camera, to which I was disappointed that I can't get for rent till later in the year.... So, I went and bought two 7D's and I LOVE them!

I'm looking at getting a side-by-side rig from Jasper Engineering (Jasper Engineering: Twin Camera Bar).

Here's a few shots we did during our testing period to determine convergence ranges.
3D - a set on Flickr

If anyone has any hints and such let me know... I've never done this before.

Thanks

Tony

I've been using 2 7D's as 3D 'training wheels' for a few months now.

The Inter Axial distance on a rail/bar is right at about 6", which is a little over 2X your Inter Occular distance between both eyes. So using the 30:1 formula, I try not to feature shooting anything closer than around 15' from the cameras. I use two sets of matching lenses, 28mm f2.8 and 50mm f1.4. I find that shooting outside I need to use a .6 ND for proper exposure, as I like to shoot at 200 ISO and f22 isn't closed enough on a bright sunny day.

Also, it REALLY HELPS to use some kind of a slate mark at the head of each camera record roll for post sync. Don't expect to be able to use the footage with sync sound though as the the cameras are not capable of being genlocked and they will drift somewhat. All in all, I'm happy with what I have learned from using them.

Lastly, I recommend trying out Tim Dashwood's 3D Stereo Toolbox software for FCP if possible. This little plugin taught me more than I can express about how to properly shoot 3D footage through trial and error.

Bruce,

Thanks a ton for your post, most helpful!

I just ordered my bar and will shoot some footage this weekend, hopefully... it's supposed to snow....

I have used FCut in the past but ordered ADOBE's master collection to give it a test drive, being that I'm very comfortable with ADOBE's other apps I assume the learning curve will be brief.

Tony

How would you recommend using the Canon 7D's or 5D's for professional 3D capture?

Even though the two camera's will not be perfectly in sync with each other, are you capable of getting them relatively close in post?

I'm planning on purchasing a pair of Canon T2i/550D to see how well they work for professional work. I hear the 7D and the T2i's video capture are almost identical in quality.

You sync in post, use a clapper.

The 7D is fine, more than fine, for professional use with 3D.

Get a fast lens (2.8 fstop or faster) and do plenty of testing to find your convergence point.

Tony

Hi

Excuse me but none of the not-syncable camera are most definitely NOT "fine" for professional work, because any sort of movement produces false depth. Rolling shutter twice as bad - that is why we use one camera upside down even on synced cameras that have rolling shutter, to have the rolling the same direction. All these artifacts are very visible and produce unnecessary eye-strain, pleases defer using such setups for paid work - for testing and getting the feet wet, for the price why not....

-Kaspar

Kaspar is right, Canon 7D's for 3D cannot be construed as a professional solution. All of the things he mentions are correct as well as a certain amount of 'retinal rivalry' which occurs when the cameras get more than a few frames out of sync. I've found that the best method of shooting with them is to either shoot very short bursts or long nature shots - think mountains in the distance with some trees pretty far away not moving too much.

I use the 7D's as training devices, I recommend looking at them this way for 3D work and in that regard they are a very cost effective way to learn 3D do's and don'ts.

For sync sound and pro use I use a beamsplitter mirror rig with cameras that can be genlocked and timecode synched. That is the only way at this point to insure perfectly matching footage.

Sorry, but would not using one camera upside down and one straight up make the rolling go in the opposite directions?

Hi

if you use 50/50 mirror both x and y get flipped so we flip the x back manually and deal with the y in post (or in camera for SI)


-Kaspar

Oh, with a mirror. OK, that makes sense then.

I guess I didn't think of this, but I guess I'll see how mine comes out.... Good thing the client already paid.. :-)

So, if i understand this correctly, if I only use the 7D's for quick shots that last for maybe a few minutes. The frames won't be so out of sync that it produces severe retinal rivalry?

Is there a workaround that can deal with the rolling shutter problem. I've read on other forums that many cameras suffer from this, maybe not as severely as the 7D's but it's fixable. Maybe Kaspar can shine some more light on this topic.

Retinal rivalry is bad whether it lasts hours or seconds.

I totally agree, I wouldn't want to produce any sort of retinal rivalry whatsoever. But from what I'm getting, it sounds like even though the two cameras can't be genlocked and if your only shooting for 1 or 2 minutes per shot, the frames won't go so out of sync that will produce retinal rivalry. Or did i get that completely wrong?

Best,
V

If they're not in sync your eyes will be suffering from retinal rivalry. If you want to see this firsthand start your cameras and then take a flash and point it at both lenses. When you play back your footage from both cameras you'll see how the shutter is in different places on either camera during the flash.

There is always a small chance that your cameras would be pretty close to being in sync on start up, but there's the same possibility that your cameras would be completely out of sync. Anyhow take that random starting time and realize that if they were traveling at the same rate without drifting then you may be able to adjust the cameras slightly in post with some plug in I've never heard of. But they wouldn't be traveling at the same rate because the cameras drift micro/milli seconds which compounds your retinal rivalry.

How do you fix this? You feed the cameras a signal to lock to and it solves your retinal rivalry, unfortunately there is nothing that will do this yet for the 7d cameras. Maybe in the future it will happen but not now. I think when you see it you'll realize the extent of the problem. Or perhaps you won't consciously realize it but you will have a splitting headache after a couple of minutes.

On the other hand I believe that enough people do attempt this and it catches on, we can train our brains to not need a genlocked source. But I have absolutely nothing to base that on except my own theories.

If they are out of sync, they are out of sync. It does not matter how long or short a shot is. They do not get the more out of sync the longer you shoot.

It is not that the two cameras do not shoot at the same speed, they do. The reason they are out of sync is that they do not start at the same moment. Let us say, for example, that you are shooting at 25 fps. One camera starts at one moment, the other starts 1/50 s later. That means they are 20 ms, or half a frame, out of sync. They are that way right from the start and they continue to be that way whether you are shooting for a minute or for an hour. If there is any kind of movement, especially lateral movement (that is left-to-right or right-to-left movement), objects will appear at a wrong depth. You can have a body 2 meters in front of the camera but its arm can appear 1 meter or 3 meters or whatever other distance from the camera. Or the body can be walking at the side of a pool of water but appear to be walking on the water, even while its feet are clearly touching the firm ground. This causes disorientation in the viewer, as his brain is receiving conflicting information.

For proper stereoscopy both cameras have to run perfectly synchronized, so this does not happen.

Yeah well.... I'm shooting and it's coming out pretty nicely. I'll post samples when I'm totally satisfied with the subject matter.

That is true Adam. I think what Tony was trying to say was that two cameras will never run at exactly the same speed therefore after a period they will stray. Genlock is esential.

NOw I only own one EX3 but I do own two HC5's ( small little consumer HDV cams) They can't be genlocked obviously, but you can by the Colorcode lanc controller.
LANC and this should start the two cameras off at exactly the same time. Obviously no Genlock so after a while the frames will not be accurate so short clips are important to avoid straying.

Am I right?

I don't believe that using LANC, in general case, one can achieve good enough frame synch in each instance. At least I couldn't do it. I have a clip here: YouTube - LANC Sync Experiment - 3D Stereoscopic HD . Each device implements the LANC protocol at a different level. There is just no way to guarantee that the camcorder controller S/W will respond to serial protocol command and control piece of H/W with the exactly same delay. That would depend greatly on what other real time tasks they have to serve (RTC, display control, switch scans,...) and how the code is written. Furthermore, due to the battery voltages being slightly different and crystal clock and other oscillator tank component tolerances, the controlling clocks will not be exactly the same, so there will be a longer time drift. IMHO, the only way to assure acceptable synch is by H/W re-sync at each frame - genlock. I know that there are the LANC Masters, etc. on the market, but I don't see how they can GUATANTEE the camcorder response in each case, since the S/W and LANC command implementation in each camcorder can and does vary. Often there are slight differences in the same model. These devices can only command the cameras at S/W level, over the b-directional LANC I/F, not at the shutter H/W level, that actually determines what the sensor can see. I would sure be interested about how can that be achieved with the LANC in general case. It can be "good enough" for slowly moving subjects, while not moving the cameras, but in general case, there can be a lot of motion within a frame or field, which will result in different shot in L/R views, in general case.

that is where shooting 60p become interesting. the frame rate is fast enough to make the eventual difference negligible.

I was replying to Bicky in that message, something about being safe with short shots. Maybe I did not word it right. My point was that even if the two cameras did run at exactly the same speed, they still need to be synchronized (yes, by genlock) or else you can get some very strange results. Even a few milliseconds off sync can cause serious issues.

Thanks for this bit.

Thanks for all the replies....most helpful.

What are the smallest and cheapest cameras out there with Genlock?

That would be worth trying. I know that Sanyo makes couple of consumer grade 60p camcorders. Can you tell me what other models are there? Does Canon make any? What editor and stereo compositor S/W can handle 1080/60p footage? I would buy the Panasonic 3D rig, but it costs almost as much as my wife's car. If I bought it, I would end up with Tiger Woods in some sort of 3D rehab and divorced. The lawyer would then get himself the Panasonic on me. ;-)

the sanyo FH1 i use for this purpose (3D) are 1080p60, but 720p60 is ok too and is found in many cameras and DSLR and is compatible with bluray (1080p60 is compatible with almost no support).
720p60 3D give 120p while interlaced. many screen and projectors already support this frame rate.
And you can still drop half of the frames and get a 3D at 60p (2x30) that should be viewable.
I think the good thing with 60p is since the pricture is progressive (no upper/lower frame), there is a big chance that you can sync to the closest frame with pretty good accuracy.
at worse if a left frame start in the middle of the right one, you only get 1/120 difference.
if it start just before or after, by aligning to the previous or next frame, you get much better than 1/120.
in my case , i have seen no problem with my FH1.

http://www.giroud2.com/download/test_left.avi
http://www.giroud2.com/download/test_right.avi

Hi

The problem is not sync drifting but inter-frame sync, meaning the sync difference less than 1/24th of a second for DCI cinema. This is enough to introduce serious problems.

-Kaspar

The trick with LANC controllers is to switch both cameras on at the same time. Assuming the cameras are identical and are equipped with the same firmware, the sync generators in both devices should reset and start aligned. By comparing the LANC bursts of both cameras (which are synced to field transitions) the controller can measure and display the field offset of both cameras. By delaying the power-on sequence of one camera, fine tuning is possible.The problem is that the startup mechanism is analog and depends on battery power, temperature and circuit tolerances. The sync drift after power-on is relatively small and depends only on the accuracy and variations of the crystal oscillators in both cameras.

There are no practical reasons, why cameras with genlock option are relative expensive. With modern chipsets this functionality could easily be implemented in compact prosumer cameras.

I will have to look at the consistency of the sync shift after the power-up, based on your input. How do you precisely delay the boot sequence of one of the cameras? How do they keep long term sync?

Exactly! The two cameras must be 100% synchronized at all times. The only exception to that would be if nothing in the scene is moving, but that kind of defeats the purpose of using video cameras.

If you can program a micro controller to control both cameras via LANC, the following sequence can be used:
(0) Power-on both cameras
(1) Send a power off LANC command to both cameras
(2) Power on both cameras by grounding both LANC Lines for 100ms
(3) After release both cameras will turn on
(4) Measure the offset of the LANC packets of both cameras to calculate the requred offset

Repeat the same sequence and apply the calculated delay

The diagram may clarify the sequence.

Hello Heinz. Thank you for your explanation. I understand how you start the cameras at sync with one another, by manipulating the power control, so that the communications bursts, which are alligned with the sycs end up alligned with one another as well. I am not sure how is the long term drift controlled, but I can see how the initial allignment is achieved. I saw a paper some time back, that showed a long term control, but I was not sure about the applicability to all camcorders. You are no doubt familiar with this: http://dsc.ijs.si/3DLANCMaster/Files...ster_paper.pdf . The simple thing that I hacked cannot achieve those results. Thank you for your great clarification. Your method is certainly very clever.

I really don't understand what results all of this LANC discussion for 3D is supposed to yield. If the two cameras are not genlocked together so that one camera or an external sync generator drives the clocks in both cameras simultaneously throughout the shot duration, then starting the cameras at exactly the same time is a moot point. LANC is not genlock. Besides, there is a much easier and cheaper way to derive a common head sync point which is almost 80 years old - use a clapper slate, or a still camera flash, or clap your hands at the beginning of a roll and use your NLE to move both shots to that exact frame. Now at least the first few frames will be in dead sync, but the inevitability of drift will appear shortly thereafter. That drift may or may not be noticeable at first but will be over time - how much drift is dependent on the cameras used but is not a mathematically predictable event that can be adjusted later. The drift will be asynchronous and variable from camera to camera. I have been shooting video (now digital) cameras since the 1970's and we have ALWAYS had to genlock two or more cameras together to achieve dead sync. Nothing has changed, not the basic principles nor the techniques - the internal clocks of all of the cameras must be driven by a common sync reference whether it be black burst for standard def or TriLevel for HD.

One can will and wish it were otherwise so as to enable more inexpensive solutions like DLSR's to work, but it's not going to happen any time soon that I can see. The reason is simple and obvious - all of these camera manufacturers make higher cost cameras that have the genlock function and don't want to see a lower-end sibling siphon off profits. Cynical maybe, but it's a business model that I have watched all of the camera manufacturers employ for as long as I've been in the biz.

Listen, I use two Canon 7D cameras to do non-pro 3D work, mostly of natural landscapes and if you are careful with your settings, it can turn out OK - not perfect but viewable. But for sync sound or any professional work, you must use two cameras that are in guaranteed dead sync, and the only way to do that is with genlock. Interestingly or ironically, that is why genlock was invented.

Clapper will not synchronize the capture within a vertical sync period. Half the frame time (16.666 ms) can result in the frame capture to be too far apart, in some instances. Since the vertical syncs are free running, if one manages to power up the two cameras in a scattered fashion so that the syncs start at nearly the same time, the sync skew will remain within and acceptable tolerance for several minutes based on the tolerance specs of the crystals and the fact that the cameras are running at a pretty close temperatures. Since the comm bursts back from the cameras are slaved on the vertical syncs, one can measure and compare their return channel timing and synchronize the cameras quite accurately by controlling their power up command precisely. For limited time duration, as well as genlock can. The paper I refered to can actually shows how in some cases this can be done for a very long time, if the phase lock loop scaller can be controlled in small enough steps. The skew can be monitored and signaled back to the controllers, so based on the desired treshold, the operator can re-sync and start again, if the skew gets unacceptable. If you wish to be mathematical about it, even genlock will have minute delays and skews between the cameras, but usually tolerable for most standard video work. The sub-frame accuracy is very important in the case of fast motion of either the subject or the camera, and miss-alignent is very obvious to see. If you don't capture fast moving objects, cars, sports, and have the cameras on tripod or some other fixed support, clappers are fine. Some of us want to get the most of the technologies, even though we don't want to deal with the cost or bulkiness of genlocked gear, so interestingly someone came up with rather clever solution, that is good enough for what some of us do. There is enough interest that there are several products solving this issue on the market. In general case, consumer cameras have no need for multi-camera sync, so adding genlock there is just a waste of profit for the manufacturers. Since some of us are interested in sub-frame sync of consumer grade cameras, we discuss it in this forum.

The name of this thread refers to using two Canon 7D cameras for 3D and that was what I was referring to. You are discussing in very good detail how to do this with a couple of camcorders, and I appreciate the depth and breadth of the discussion, but I haven't seen any method to use that process with two DLSR cameras which are not equiped with that protocol.

It's an interesting theoretical discussion, I'd be interested to hear and see real world results of it's application.

Probably the Cannon XH-G1s, about $7k. Also the JVC GY-HD250U, also about $7k, no lens. Both shoot HDV to miniDV cassettes. Both have SDI out if you want to use a better recording system.

you might be able to find some sort of HD box camera for around the same price with genlock in and SDI out. A box camera would not have any built in recording, though.

Small & cheap don't exist in the same product! The smallest genlockable HD cameras available are the Iconix, Cunima and SI-2K mini, but none of these have built-in recording. They are just the heads.

HI Tim, I thought that was the case!

I am playing around with a couple of consumer Sony HC 5's and must admit the results are quite encouraging. I don't like the 1080i the cameras use but I am relatively pleased with the results. For test purposes only of course.

Might see if I can pick up a pair of Canon HV 30's as they shoot progressive.

The PMW-10MD has genlock, this is a very samll "C" mount 3 chip remote head (1/2" 1920x1080 sensors) that connects via an umbilical to a control box/recorder that records on to SxS cards. Not cheap though, around 12,000 Euro !!

If you get a pair of Canon's or Sony's with Lanc, getting them in reasonable sync is pretty easy.

Alistair, what do you tink of the Colorcode 3d lanc control?
http://www.colorcode3d.com/LANC.html


Worth the money?


I have a couple of HC5's and a HC3. I might either buy another HC3 or flog all three and get a couple of HV30's.

It would be nice to have higher end cameras for 3D ( and I will be buying another EX3 to go with the one I already have) when I need it.

The results I have got with a couple mof HC5's are excellent considering its interlaced single chip HDV. Not good enough for any commercial work but good enough to learn how to make good 3D, and they are very light.

Hey guys,

I just wanted to double-check something about using the 7D's for 3D. I understand they can't be used correctly for 3D, unless the image is static.

But I remember reading in a couple of places that if you shoot in 720p @ 60fps that you can possibly shoot quite effectively with the 7D for 3D which includes motion.

I was hoping you guys could confirm if this is possible or still not acceptable.

The 3D content I plan to create will only be displayed on 3D monitors and/or 3DTV's. Not for the big screens. I was wondering if 3D content is more comfortable to view on smaller screens even if the 7D's aren't perfectly synced that may cause retinal rivalry, considering they are shot in 720p @ 60fps.

This is my last hope to use the 7D for some 3D projects before purchasing an SI-3D system.

Looking forward to hearing from you guys.

Best,