Back to the main subject: the 3D look. For me, some lenses just have it, and others just don't. Below are two very poor photos taken on two different lenses. One was taken in interesting light and is under-exposed. The other was taken under very ugly fluorescent energy saver bulbs and is very underexposed as well as having terrible colour balance. Both lenses were wide open.
The right hand lens routinely produces 3D pop even in the most unlikely circumstances. In this case the shutter speed was well below the handholding limit, the picture is shaky, the iso is too high, the colour balance foul, and everything is wrong. It has no sharpness or micro-contrast, but weirdly, loads of 3D. The left hand lens struggles to produce any 3D effect at all even in circumstances where it should.
For me it is definitely nothing to do with sharpness and all to do with realistic presentation of light and shade.
Richard, a 15 cm diameter would yield a head circumference of 47 cm, which would be highly pathologic for a newborn. It would be around the 50th percentile for a 15-month old, but probably 4 or 5 standard deviations above the mean for a young infant. It's not quite so straightforward, because the head isn't symmetrical, but you get the point -- 11 or 12 cm is much more likely for the longest head dimension in young infants. Individual features on the face will probably be of more interest than the extremes of the head, though, and these features are a cm or two apart and a cm or two deep.
Here are head cirumference growth curves that we use in practice:
I do think that sharpness can have a tremendous effect on 3D effect. Having abundant textural detail is extremely important -- but of course you need permissive lighting to see the detail.
This, by the way, shows quite well the types of bokeh and long transitions of focus you can get with large format. This is on 4x5 using an uncoated, 35 year old Schneider symmar lens. It was taken with dull, flat lighting, but I used Velvia which exaggerates contrast.
Richard,
I think the best proof of movement is in your difference picture (I bumped the contrast). The fact that some but not all of the vertical elements show over a pixel difference...is "proof" enough for me that Jeff moved the camera. We can argue the math more, but I think that's moot.
I must say i agree with those who say the 3d look has not so much to do with shallow depth of field, it does obiviously provide some sense of depth.. but lets keep qualitfications seperated..
I stumbled across this unusualy sharp image on the DP forums (i know i know.. i really dont know what to do with my time...) PLEASE have a look!
Grant808 wrote:
Richard,
I think the best proof of movement is in your difference picture (I bumped the contrast). The fact that some but not all of the vertical elements show over a pixel difference...is "proof" enough for me that Jeff moved the camera. We can argue the math more, but I think that's moot.
Grant, when you 'bumped the contrast' you effectively made a quarter of a pixel of parallax look like a whole pixel. But as I have shown, even if there were one pixel, as you incorrectly suggest, and even if the AP distance were 10cm, we still get a ridiculous shooting distance of 500 inches, or 40 feet.
In fact, what you have revealed in the diff is the slight mismatch in pixel pattern produced by a tiny rotation of the camera, not a sideways movement. How do I know this? I know because if you remove the pixel pattern from the equation by uprezzing both images bicubically so that image data is the only ingredient, a diff shows even less parallax:
There is effectively no difference between these two pictures, except that the whole image is shifted across the pixel grid by half a pixel or so. That is not parallax. That shift is what your exaggerated diff is, well, exaggerating. The diff I show above, without exaggeration, is the true diff between the two images. To get a diff like this, handholding would not be good enough. You would have to put the camera on a tripod and be very careful not to move it. At all.
In other words, the differences which you have exaggerated in your higher contrast diff arise because the camera swung round by about half a pixel between these two frames. That is a tiny amount of camera rotation. Since the pixels have discrete levels, the diff will reveal areas where they do not match exactly. In this case the mismatch is not due to parallax, but camera swing. To overcome this you can uprez, which allows you to align the two images on top of each other with 'sub-pixel rendering', i.e. it effectivley allows you to move the top pixel across by half a pixel, and thereby line the image up more accurately than was possible with the cruder pixel pattern.
Regardless of the diff discussion, there are 107 lateral pixels between the rear forearm behind the head, and the top of the left hand in the foreground IN BOTH PICTURES. That shows there is no parallax. No parallax. It's incontrovertible.
Edited by brainiac on May 10, 2007 at 08:25 AM GMT
brainiac wrote:
There are 107 lateral pixels between the rear forearm behind the head, and the top of the left hand in the foreground IN BOTH PICTURES. That shows there is no parallax. No parallax. It's incontrovertible.
Could the problem be that he shifted the camera 5 inches without rotating it? So the sensor-lens-subject axes were parallel in both pictures. Thus, he might have almost employed a stitched-pano type of technique, which would cause the spatial relationships to remain nearly identical. I've never done this before, but I'd think that you'd want to move laterally, then rotate so that the center of the frame was occupied by the exact same subject.
Could the problem be that he shifted the camera 5 inches without rotating it?
No. Parallax will appear whether you rotate the camera or not. Parallax always appears when your viewing position moves sideways. Think of the lined up lamp posts. It is impossible to move sideways without opening up the gap between them, whichever part of your frame they are in.
There was no sideways shift of the camera between these two frames BECAUSE they exhibit no parallax. Like I said, it's incontrovertible.
BTW, if anyone is unsure how these diffs are generated, you can check it for yourself like this:
save Jeff's stereograph
open in photoshop
use image size to uprez the image - I recommend at least quadrupling the dimensions
select a marquee over one of the two pictures
float it onto its own layer
in the layers pallet set the blending mode for the top layer to 'difference'
select the move tool
move the top layer over the bottom image
zoom to 100%
use the arrow keys to align the pictures
the more accurately aligned the pictures are, the fewer non-black pixels you will see
once it's perfectly aligned (i.e. pretty much all black) you can now downrez it to its original size
the blackness of each pixel is the measure of the extent to which it differs from the one that lies beneath it
the fact that these images produce an almost perfectly black frame shows that there are no appreciable differences between them and therefore they must have been taken from the same position
Okay, Richard....I stand corrected. With a 200% uprez, I barely see any difference at any level.
I didn't see it with your sample counting 107 pixels because of the shift/swing between the two shots, and I couldn't absolutely tell if one was really 106.5 and the other 107.5. The uprez should have taken care of that.
I didn't see it with your sample counting 107 pixels because of the shift/swing between the two shots, and I couldn't absolutely tell if one was really 106.5 and the other 107.5. The uprez should have taken care of that.
Exactly. In fact, if you look at the blue lines in both crops from the right hand image, you can see that they are both shifted about half a pixel to the left compared to the first image. I was very careful to draw both to the left of the half pixel position in each case so that the parallax measurement was accurate to the pixel. As you say, uprezzing would have allowed effective half-pixel measurement.
brainiac wrote:
No. Parallax will appear whether you rotate the camera or not. Parallax always appears when your viewing position moves sideways. Think of the lined up lamp posts. It is impossible to move sideways without opening up the gap between them, whichever part of your frame they are in.
There was no sideways shift of the camera between these two frames BECAUSE they exhibit no parallax. Like I said, it's incontrovertible.
Understood.
Although as you'll know if you use view cameras with rear movements, rotation (i.e. swing) of the film plane will change the perspective relationships. This is basically the horizontal corollary of perspective convergence. So even if parallax exists in both scenarios, there will be perspective differences that arise due to rotation.
Sure - but if your lens and the two lamp posts lie on a straight line, then no amount of moving the back will show a gap between the lamp posts. Only shifting the lens sideways can do that.
I have to admit that the above picture has a much stronger 3d effect than many of the other photos. For some reason the girl in teh back ground just seems to float in the background. Is this my imagination ?
Well, there's a wall of mist between her and the background.
Remember that in landscape photography atmospheric haze will accentuate the sense of distance. In the case of this photo, the entire background already looks more distant thanks to the DOF, but the mist makes them look yet farther away.
So I think the photo is 'built' to have a 3D effect, most of which is probably not attributable to the lens' own features.
While light and DOF *can* accentuate 3D effect, I still maintain that the lens and the lens alone can offer a 3D effect regardless of much of what has been discussed thus far. This evening I will post some infinity shots (with no haze or fancy lighting) that have dramatic 3D effect. It will be obvious at that point that the only source of the effect is the lens (oh and the scene itself).
cogitech wrote:
While light and DOF *can* accentuate 3D effect, I still maintain that the lens and the lens alone can offer a 3D effect regardless of much of what has been discussed thus far.
And the opposite can also be true. Which leads me to believe that a 3D effect is not a single phenomenon as much as it is a reaction that we have.
Here's a challenge for everyone. Post a 3D-looking image taken with a crappy lens. Better yet, a 3D-looking image, taken with a crappy lens, and with infinity focus. I've got plenty of my own I can share.
I mean, I took these pictures using the Canon 18-55 kit lens -- it doesn't get much cheaper than that.
Assuming there is one such thing as a 3D look, all of the exceptions out there suggest the following to me:
A good lens is helpful, but neither necessary nor sufficient.
Shallow DOF with good bokeh is neither necessary nor sufficient.
Image sharpness is neither necessary nor sufficient.
In fact, no one thing is necessary, and no one thing is sufficient. The convergence of several factors, which can include effective lighting and/or differential focus and/or differential color and/or high textural resolution and/or microcontrast can create a 3D image, and it appears 3D because we see in the image the normal clues that in real life suggest depth.
