Hehe, I think Paul and Richard are having their own personal 3D effect.
As to the question, I actually don't think the image has an especially evident 3D effect, although I don't think that is necessarily a negative attribute. In other words, the presence or absence of a 3D effect is not requisite for a picture being successful or even great.
This portrait has fairly uniform lighting throughout, and because of that the 3-dimensionality is not emphasized (even in contrast with the background). Higher contrast, angular lighting might make it more 3D. Also, because the background is just a flat backdrop, the model is taken somewhat out of the context of a 3D world and I think that also makes it look less 3D.
brainiac wrote:
So either Jeff has used two shots taken from the same place, or he was photographing from at least 80 feet away. Correct me if the maths is wrong...
I don't think it requires that he be that far away unless he is using a very long lens. As I understand it this is also an effect of the parallax difference as a proportion of the total arc. So whether you have a subject 10, 20, or 100 feet away, moving the camera x inches will be a different proportion of the total angle of view depending on the focal length of the lens. But I'm no expert in this area (or any other, for that matter).
DrPablo wrote:
Hehe, I think Paul and Richard are having their own personal 3D effect.
As to the question, I actually don't think the image has an especially evident 3D effect, although I don't think that is necessarily a negative attribute. In other words, the presence or absence of a 3D effect is not requisite for a picture being successful or even great.
This portrait has fairly uniform lighting throughout, and because of that the 3-dimensionality is not emphasized (even in contrast with the background). Higher contrast, angular lighting might make it more 3D. Also, because the background is just a flat backdrop, the model is taken somewhat out of the context of a 3D world and I think that also makes it look less 3D....Show more →
Excellent assessment, IMHO.
I might also add that a part of the 3D effect, for me, at least, is white balance and how "natural" lighting looks. In other words, I have *noticed* the 3D effect most on naturally lit subjects with correct WB.
DrPablo wrote:
I don't think it requires that he be that far away unless he is using a very long lens. As I understand it this is also an effect of the parallax difference as a proportion of the total arc. So whether you have a subject 10, 20, or 100 feet away, moving the camera x inches will be a different proportion of the total angle of view depending on the focal length of the lens. But I'm no expert in this area (or any other, for that matter).
I measured the parallax shift in millimetres at the subject plane. That means the numbers are entirely independent of whichever lens you use. Half a millimetre of parallax or less over 10cm depth at the subject plane and 5 inches of pupil separation means an 80 foot shooting distance, irrespective of which lens you use. Look at the diagram; it's quite easy to visualise: http://cyberphotographer.com/5D/parallax.jpg
But if your lens has a 20 degree field of view, then any amount of movement along an arc will comprise a 5-fold greater movement (as a proportion of the entire angle of view) than if your lens has a 100 degree field of view. This is assuming, of course, that you have the same subject distance for both scenarios.
So the amount of separation you're measuring will remain constant in an absolute sense, but it will comprise a larger or smaller sweep within the arc of the lens depending on its angle of view. Right?
brainiac wrote:
I measured the parallax shift in millimetres at the subject plane. That means the numbers are entirely independent of whichever lens you use. Half a millimetre of parallax or less over 10cm depth at the subject plane and 5 inches of pupil separation means an 80 foot shooting distance, irrespective of which lens you use. Look at the diagram; it's quite easy to visualise: http://cyberphotographer.com/5D/parallax.jpg
The biggest damning factor here is that you are using guesses and bad math to "prove" to yourself that I am lying, when I am not. You can trust me when I tell you that I moved the camera several inches, and that I was not 80 feet away. You're wrong.
I think this has all been prompted because you feel that your 3D image is far superior. You cannot see the 3D effect in mine, which comes as no surprise if you think that yours looks normal. People don't normally expect to see an inch or so of neck disappear from the view of one eye; this is all I was talking about. If you are used to looking at such overdone 3D then a normal one may well look flat to you.
Jeff, do you have other Stereo photos to share?
I can not see the Stereo in your baby shot, nor do I see it in Brainiac's, though I get a headache of his pictures, so there might be something there.
Maybe some other pictures would make it clear to me.
Jeff, I am not accusing you of lying and I certainly don't think my portrait is superior to yours in any way. I did it casually as an enjoyable experiment. As I said when I first introduced my stereogram, the 3D is exaggerated because I couldn't get the cameras close enough together. This is not a competition.
I think you made a simple error that any one of us could have made. There is no parallax shift whatsoever between your two files. Have you looked back at your originals to see if there is one that shows a parallax difference against these two?
> But if your lens has a 20 degree field of view...
Paul, if you shoot with a zoom lens, as you zoom in from each position you will not see parallax change at all. In other words, line up two lamp posts and zoom to 24mm. Now zoom in to 85mm. A gap will NOT appear between the aligned lamp posts as you zoom in. Take one step sideways and the gap WILL appear. Zooming only changes the gap's width in your viewfinder, not the gap's width in proportion to the lamp posts. Parallax shift derives entirely from your shooting position, and not in any way from which focal length you are using. That is why no knowledge of focal length is needed to work out the shooting distance. It is a result of simple Euclidian geometry as this diagram shows: http://cyberphotographer.com/5D/parallax.jpg
The image has about 2 pixels per millimetre of subject, so one pixel or less of parallax (as illustrated by the photoshop difference layer) equates to 0.5mm or less at the subject plane. We can see from the subject that the left hand must be 100mm or so in front of the right arm. So parallax=0.5mm, depth=100mm, pupil separation=5 inches and therefore shooting distance must be at least 1000 inches, irrespective of which lens was used.
Since a shooting distance of 1000 inches seems wrong by an order of magnitude, we have to question one of the other measurements, by an order of magnitude. The only one that has a chance of being wrong by that much, is the separation. The pictures are effectively identical, as the diff shows, so it seems Jeff has accidentally used two images shot from the same position rather than two from positions 5 inches apart.
Apologies for labouring the point, but it is an interesting technical issue that's worth understanding and it lies at the heart of 3D vision so it's not even OT!
This may or may not be related, but I've always regarded the susceptibility of long lenses to motion blur to arise from a greater ratio of motion arc to total angle of view. The difference in that case may not be parallax, however, but rather the difference in reproduction ratio on the film / sensor.
Parallax has little to do with focal length exept as related to the distances from the different objects in a stereo composition. Jeff's picture doesn't have much pixel separation because the items you are comparing are very close to each other.
Except that there are different degrees of binocular convergence with human vision depending on the subject distance. When you examine a patient and have them focus on objects at different depths, the angle of convergence between their eyes changes (look at the tip of your own nose, for instance). So the human visual system uses convergence to account for parallax with subjects that are at different distances.
Of course the human eyes have a (relatively) fixed focal length, and they don't physically shift closer or farther apart.
brainiac wrote:
1/2mm of parallax between objects 10cm apart gives a ratio of 200
therefore ratio of eye_separation to shooting_distance is also 200
i.e. shooting_distance = 200 x 5 inches = 1000 inches = 80 feet
So either Jeff has used two shots taken from the same place, or he was photographing from at least 80 feet away. Correct me if the maths is wrong...
Okay...correcting the math:
Say your assumptions are wrong and the separation is 5cm and the eyes are now 3" apart...I get a shooting distance of 25 feet. Makes sense to me.
I think I may have misstated before about the 5". I remember measuring my eyes and then using roughly that measurement. I often remember numbers as evens or odds. If it helps I may have misremembered 3 as 5, I don't know. Sorry if that's true.
I have an important exam coming up, but on Thursday will recreate the shot with a similar-sized object. 25 feet is roughly what it could have been, although still just a little long; i used a zoomed-in 24-70L in my living room on a tripod. Of course, there are lots of assumptions and eyeball measurements in the math; for instance, "half a millimeter" is not an actual measurement.
And to correct myself...looks like the average eye separation is about 2.5" or 63mm. Which would take the previous example to 21 ft...
Even the change in assuption from 10cm to 5cm for the front to back distance is probably way off...so a shooting distance of something under 21 feet is likely.
Just trying to get the right order of magnitude...and not make a huge case of propogation of errors.
Grant, a baby's head is more like 15 cm anterior-posterior not 5cm. I suggested 10cm just to give Jeff the benefit of the doubt. When I offered one pixel of parallax I was being very generous too. Forgive me for trying to explain it another way:
I have chosen the point where the forearm goes behind the head so there is no way it can be before any part of the head. I have chosen the top edge of the hand since there is no way it can be behind the head. http://cyberphotographer.com/5D/stroke.jpg
My estimate as to the distance between these points along the lens axis is 15-20cm but I accept that if anything that is too little.
Along the top of the images we see the x-axis pixel references for those points. 328 - 221 = 107 correct me if I am wrong. 759 - 652 = 107, correct me if I am wrong. Therefore there is no parallax difference between these two images and they are both taken from the same place.
Even if we imagine that this measurement is one pixel out, then the following holds:
Therefore each pixel is about half a millimetre across at the subject plane. So one pixel of error gives us a ratio of 0.5mm parallax to 150mm subject depth, i.e. 300:1. Even if the pupil separation were 2.5 inches, we are still looking at a shooting distance of 300 x 2.5 inches i.e. 750 inches. With the 70mm lens that Jeff used, we must be looking at something like a 100% crop.
Jeff, please put us all out of our misery by finding the picture that WAS taken from a different position. And good luck in your exams. I mean it.
Edited to include blue circles picture for clarity.
Edited by brainiac on May 09, 2007 at 08:04 AM GMT
10 cm is the closest. A typical head circumference for a healthy newborn is 33-35 cm, which would make the diameter about 10-11 cm. (yes, most babies fall in a Gaussian distribution with a broader range of "normal" than that, but these are typical values). The anterior-posterior dimension, from the forehead to the occiput, is longer, because babies don't have spherical heads; and vaginally born newborns often have molding (i.e. misshapen heads) for the first few days after birth because of compression during delivery. Older babies, of course, have a larger head circumference.
So Paul, in your professional opinion, would you say that 15cm is a likely depth for the baby's head? And therefore would you agree that 15cm is a conservative estimate for the axial distance between the two points I have used for parallax measurement?
