How accurate is pixel scaling?

[Enlightenment]

I wasn't really sure where to post this but given that it's marginally related to the 'Adarx & Mughi called out' thread I thought here was the best place.

I've been wondering about the real accouracy of pixel scaling for some time and with the ongoing Adarx/Mughi smackdown on SB now seems like a good time to post some tests in an attempt to test the usefulness of this method.

Exhibit one: http://www3.sympatico.ca/gcircle/pix-scale-1.jpg

This exhibit uses square pixels and the Executor mesh has been scaled to 18.289km in length. How large is the ugly work-in-progress that's sitting directly below the Executor?

Exhibit two: http://www3.sympatico.ca/gcircle/pix-scale-09.jpg

This exhibit is the same as exhibit one except that it was rendered for an output media with non-square pixels, such as NTSC D2 widescreen. How large is the second ship in this case?

I'll post the actual size figure for the ugly ship once a few estimates have been posted.

Credits: Executor mesh by jarjarthomas of http:/./www.novaimages.de. Conversion to Lightwave by Jason Tinsley.

[Darth Wong]

It is impossible to tell, because neither ship is behind the other, and there are no independent reference markers to give us an idea of relative distance. The ugly ship could be 200 metres away from the camera and miniscule, or 50 km behind the SSD and gigantic.

What does this have to do with "pixel scaling?" The pixels are merely quantization of an image, and as long as any image is resized to the proper aspect ratio, they don't matter.

[Connor MacLeod]

It is impossible to tell, because neither ship is behind the other, and there are no independent reference markers to give us an idea of relative distance. The ugly ship could be 200 metres away from the camera and miniscule, or 50 km behind the SSD and gigantic.

What does this have to do with "pixel scaling?" The pixels are merely quantization of an image, and as long as any image is resized to the proper aspect ratio, they don't matter.

Apparently it does matter... if some moron is making an issue out of it.

What I find laughable is that Fivers are now resorting to altering methodology at will to suit their purposes for debates. Note that they'll shoot it down when its inconvenient, but may very well resort to those methods when it suits them (remember the "visuals" they've used to support such arguments like firepower, acceleration, etc.)

[SyntaxVorlon]

What exactly is this for, are you just creating pretty pictures, or are you planing to put this in some sort of release of some kind.

BTW:
Wong you can tell how big the second ship is, get out your slide rule and see how long the SSD and other ship are in the first and second pictures. Check the ammount that each got longer by in the second picture and compare that ammount, percentage wise, to the first picture, which ever got bigger faster is closer and I'm sure there are certain laws of geometry that would allow you to figure out how far the thing is from and how big it is in comparison too the SSD. In the future Enlightenment you may want to include another known to make the whole thing mure understandable.

[Connor MacLeod]

It is impossible to tell, because neither ship is behind the other, and there are no independent reference markers to give us an idea of relative distance. The ugly ship could be 200 metres away from the camera and miniscule, or 50 km behind the SSD and gigantic.

What does this have to do with "pixel scaling?" The pixels are merely quantization of an image, and as long as any image is resized to the proper aspect ratio, they don't matter.

Additionally, one might point out this is why "UPPER" and "LOWER" limits are often used, even in scaling. Much of Brian's scaling work (except for stated values like b5 or the Excalibur) often relies on this.

[Darth Wong]

Wong you can tell how big the second ship is, get out your slide rule and see how long the SSD and other ship are in the first and second pictures. Check the ammount that each got longer by in the second picture and compare that ammount, percentage wise, to the first picture, which ever got bigger faster is closer and I'm sure there are certain laws of geometry that would allow you to figure out how far the thing is from and how big it is in comparison too the SSD. In the future Enlightenment you may want to include another known to make the whole thing mure understandable.

Actually, since you have no reference points, you still can't figure it out. If the ships are different sizes in the two shots, you can only tell that each ship got closer to you by a certain percentage of its original distance. You still don't know what that distance is.

For example, let's say you have object A which is 10km away and gets 5% bigger. This would mean that it is now 9.5km away. Now let's say you have object B which is 100 metres away and gets 5% bigger. This would imply that it is now 95 metres away. But if it was 50km away and it got 5% bigger, it is now 47.5km away. How can you tell?

[Enlightenment]

It is impossible to tell, because neither ship is behind the other, and there are no independent reference markers to give us an idea of relative distance. The ugly ship could be 200 metres away from the camera and miniscule, or 50 km behind the SSD and gigantic.

Perhaps I should have been a bit clearer. The ugly SOB is directly below the Executor not just in the image but also in 3D space. The centroids of both ships differ in position only by their y coordinates (y = up).

As far as I can tell the exhibits I've posted here as a test for scaling ships based on photographic aren't that much different from the stills used by Saxton to scale the SSD in the first place.

[Darth Wong]

Oh, well then I'd eyeball-guess a little over 20km. What were you hoping to accomplish with this? The "square vs non-square" pixel thing is a bit silly, isn't it? Images displayed on a computer use square pixels; if you render for non-square pixels and then display them as square pixels, you are distorting the aspect ratio.

PS. Saxton's scaling is rather different from this case, since we have an ISD between the SSD and the camera, which allows the setting of limits. We also have the bridge tower in relation to the entire ship, which allows direct geometric scaling.

[Brian Young]

Pixel scaling is like this: http://www.babtech-onthe.net/wsb5bay.jpg
In that image, the Whitestar is between Babylon 5's landing bay and the camera. How much closer, we don't know, so exact scaling is impossible. However, if the size of this section of Babylon 5 is known, an upper limit on the width of the Whitestar can be derived.
I capture all of my images in a 4:3 ratio, almost invaribly at either 320x240 (one field) or 640x480 (both fields), so this image is in perfect perspective to what we see on our TV screens.
We can easily see that the Whitestar is not as wide as this section of Babylon 5, but how do we be more specific?
Well, we could do a gross measurement by printing the image and measuring each one, but do all printers print the same way? I don't know.
The most precise and most easily duplicated way is to count how many pixels difference there is. B5's front section is X pixels wide, and we know it is Y meters width. The Whitestar is Z pixels wide.
Cross-multiply to find a number on the Whitestar's width. The result is an upper limit. The Whitestar is closer to the camera than Babylon 5, thus, it appears larger relative to B5 than it is. Thus, it can be no larger than what it appears here. Thus, this is an upper limit

Then, we find images where the Whitestar is father away than objects of known size. We can then derive a lower limit.

Keep doing this to get more upper limits and more lower limits. This not only validates the original limits set, but usually gets closer and closer to the actual size.

This kind of scaling rarely provides exact sizes, but provides a minimum and maximum size.