New JVC 46-inch 3D Monitor Costs $7,000

[Eric Lange]

Originally Posted by istvanttt

Hmm. I just made some calculations and I think when I keep the min/max distances between 1m and 3m then I can use with the 100mm an interocular distance of 10mm which should give me an interesting close-up and not be too flat neither too hyperstereo. I think the flatness of the longer lenses comes more with a higher distance. Anyway today I bought a second Nikon D300 and will test it with the beamsplitter rig I built. I want to play with the emotions I can get with longer lenses on different interocular distances. Will see.

To the second part, yes I see and understand what you say but still cannot find help to decide. ie some months ago I asked an optical scientist who did thousands of s-3D pictures for help and he suggested: just stay away from assimetrical lenses.... hmmm

On the part of the lens threats they never talk about the problems we poor and lost s-3D kids may have with the lens choices. Unfortunatley for us the choices cannot be based only on taste and beauty but need extrem precision on the external parts of the image. To give an example to other readers. In the time of the classical SLR film cameras the acceptable parallax was considered 1.2mm (1.8 in the case of anagliphic pictures) on the slides. Now I think the strenght of the distorsions can change a lot in 1.2 mm and create undesired vertical parallax problem towards the corners.

Anyway, probably even if something in my stomach tries to say me to wait, I probably will give up for now and just buy the RED lenses. As I ordered them very early I may be even in the first 100 units. The problem will be more when will I get a second RED and the second lens-set?
Ciao
Istvan

We had technician four years ago mistakenly shoot an object at 1.5m using a 100mm lens on a 24mm sensor of an object that was 5” in diameter (an early medieval copper Jug), base separation of the order of 10mm. The jug appeared very flat, the roundness had in effect been compressed by at least a factor 4. [We have a stereo plotter so we can determine this more easily]. The apparent 2.5” radius of the object was compressed to an apparent depth of about 10mm. The fine details were not well resolved in depth either. I think if you want to do an extreme close up in stereo you are better off with a wide angle lens and an extension tube.

I think you read my mind, because I was going to say that perhaps you should buy a D300 (or two) and experiment with a bunch of cheaper lenses to get a feel for the kinds of stereo cinematic effects you want to achieve. We generally calculate things very precisely for set ups but invariably we make moderate adjustments in parallax once we view the stereograms. Wider angle lenses are much more forgiving, and longer lenses (I find need secondary adjustments). You may find that you discover an awesome cinematic effect using a 100 mm lens exploiting its flattening capability against a shallow DOF for a cool impressionistic stereo background for example. In time, you will find that you rely less on calculations and more on experience for the right set up.

Your optical scientist friend is right, like lost most photogrammetrists we favor symmetric lenses, i.e BIOGON type design because of its low distortion for metrological applications. However, there are two main trade offs with symmetric lenses. 1. Relative illumination from centre to edge has a very steep fall off; 2. Contrast ratio (and in some cases resolving power is less than a retro-focal lens). Schneider has a terrific Apo-Digitar 5.6/47mm lens (very high res) with less than 0.3% distortion that we are fitting to Medium Format digital backs for stereo photogrammetry (our aim is to take the best stereograms in the world); however we have to use a gradated centre to edge neutral density filter to compensate for the uneven relative illumination. This lens would be good for the 35mm future red sensor designs. At least you do not need to correct for radial lens distortion for every frame in post for stereo movie applications. I don’t think anyone has yet manufactured a symmetric cinematic prime, as they are all retro-focal designs. Biogon lenses (apart from ZEISS UMK) are much slower than cinematic primes and of course, do not compensate for breathing (maybe not such a big deal in stereo anyway)

In your initial experimentation don’t sweat the radial distortion too much anything that has less than 2% distortion still makes very viewable stereograms providing you don’t have to look at them all day long. Remember radial distortion does not directly = y parallax, so if you have 1.5mm radial distortion at the edges it does not mean you have 1.5mm Y parallax. The Y parallax comes from when you overlay the distorted images, one image on another with a horizontal image translation (for example an 80% overlap). If you overlay two distortion curves of the left and right lenses onto of each-other and then slide them apart so there is an 80% overlap it is the difference between the left distortion curve and the right distortion curve that gives you this residual Y parallax. Typically for a 2% maximum distortion lens, with an 80% overlap a 0.25% residual difference in distortion will be found. The pattern of Y parallaxes is of a vertical lozenge shape in the centre 30% of the image image which does not have significant Y-parallax and then it ripples out towards the edges to higher Y parallax values (It’s a complex pattern, I will post some time our MATHCAD simulations of this to better explain what’s going on). However radial distortion does cause flat planes to become bowed. Negative distortion causes flat planes to be bowed from the centre to the edges being bowed even more in depth away from the camera, i.e. it turns flat orthogonal planes into convex surfaces, with secondary more complex ripples that are indistinguishable to the eye.

Some of the things I would be more concerned about are the mechanics for example of well matched stereo follow focus (as this could get quite expensive with cinematic hardware, and not necessarily very precise. (we are trying engineer some alternatives to this approach that would be more precise and yet much cheaper).

Your gut instincts are very sound, in terms of waiting, you may even want to buy high res Scarlets (when they come out) as they seem (on paper) much better suited to stereo applications, small form factor large sensor size and so on. When you build your mirror rig you might also consider facing the camera lenses into each other and have two small mirrors at 45 degrees facing forward. You should be able to get down to a 3cm base separation (precision adjustments are required), but is smaller and cheaper than a big mirror (pellicule) rig. The cameras can be slid apart for larger base separation until there is enough clearance for the cameras to look forward (assuming a precise track slide mounted on a tilting head).

Good luck,

Hope to see some gorgeous stereograms of Venice from your D300s in the not too distant future. (we could build some nice 3d models from the stereograms you take :))

Cheers,

Eric

[istvanttt]

Originally Posted by Eric/VERAXYZ

......

In your initial experimentation don’t sweat the radial distortion too much anything that has less than 2% distortion still makes very viewable stereograms providing you don’t have to look at them all day long. Remember radial distortion does not directly = y parallax, so if you have 1.5mm radial distortion at the edges it does not mean you have 1.5mm Y parallax. The Y parallax comes from when you overlay the distorted images, one image on another with a horizontal image translation (for example an 80% overlap). If you overlay two distortion curves of the left and right lenses onto of each-other and then slide them apart so there is an 80% overlap it is the difference between the left distortion curve and the right distortion curve that gives you this residual Y parallax. Typically for a 2% maximum distortion lens, with an 80% overlap a 0.25% residual difference in distortion will be found. The pattern of Y parallaxes is of a vertical lozenge shape in the centre 30% of the image image which does not have significant Y-parallax and then it ripples out towards the edges to higher Y parallax values (It’s a complex pattern, I will post some time our MATHCAD simulations of this to better explain what’s going on). However radial distortion does cause flat planes to become bowed. Negative distortion causes flat planes to be bowed from the centre to the edges being bowed even more in depth away from the camera, i.e. it turns flat orthogonal planes into convex surfaces, with secondary more complex ripples that are indistinguishable to the eye.

......

Your gut instincts are very sound, in terms of waiting, you may even want to buy high res Scarlets (when they come out) as they seem (on paper) much better suited to stereo applications, small form factor large sensor size and so on. When you build your mirror rig you might also consider facing the camera lenses into each other and have two small mirrors at 45 degrees facing forward. You should be able to get down to a 3cm base separation (precision adjustments are required), but is smaller and cheaper than a big mirror (pellicule) rig. The cameras can be slid apart for larger base separation until there is enough clearance for the cameras to look forward (assuming a precise track slide mounted on a tilting head).

..........

Eric

Why comes it that at the end its only we two to write each other in this threats?

I do not have a good preparation in optics, so my contribution is limited. In any way, it clear that the delta of the two distorsions have to be considered. But are you sure that it get to only 0.25% of delta on the corners also with general lenses? Or you talk about your specific fotogrammetry lenses which are specific. The reason why I'm so concerned about this radial distorsion is that I saw in so many cases on clips I made that towards the borders the textures often flickered. The geomery was ok, the main objects ok, just the textures where "nervous". I had allways to crop that part. True the cameras where cheaper Panasonic cameras. I don't think they where digital artefacts becasue it was never to the right/left extrem but only towards the corners.
Unfortunatley I never keep bad materials on my HD so I cannot send you something to explain better.

Could you PM me a drawing how you can get to 3cm interocular distance with 11cm diameter 25mm (or even 18mm) cinema lenses put in front of each other pointiing to a 45° mirror set up?
Have a nice weekend
Istvan