QLT 02: Stones: Spak Nogut Tru

    By: 

Nathanael

Add to Favorites
Embed
Facebook
Report Abuse
Share

Description

This is the third in a set of three synths designed to determine whether giving the synther photos leveled with a level attached to the camera increases the number of image features that the synther will track, as suggested by Jim Cseke here:

http://photosynth.net/discussion.aspx?cat=6b63cb81-8b57-4d5d-a978-41d5509bf59a&dis=c13fd173-a14e-4edf-8350-a3c28d6fb859

To reveal the pointcloud, click on the second icon from the right and click 'Point Cloud'.

This synth is composed of the intentionally crooked photos. To compare these results to that of the casually eyeballed photos and the leveled shots, open the following synths in a separate tab and flip back and forth between them.

Eyeballed:
http://photosynth.net/view.aspx?cid=3034f0ff-0084-45f6-9688-3bc82bf2b958

Leveled:
http://photosynth.net/view.aspx?cid=fed01de7-4ad9-432b-8977-ab60f93c521c
Stats
Synthy 100%
Views 65
Favorites 0
Photos 19
Date Created 5/4/2010

Related Photosynths

Comments

(16)
jimcseke Over 1 year ago
What does "Spak Nogut Tru" mean?
It sounds kind of cool.
Nathanael.Lawrence Over 1 year ago
Ha.

That's Melanesian English/Pigin English for "completely wasted" or as the Midwestern Americans of a few decades ago might say, "drunk as a skunk".

You pronounce the vowels much like you would for "Spock no good true".

Basically, the thinking was that although I found that I generally take photos that lean a little to the left naturally, they didn't seem terribly different from the level shots. I wanted some shots that would really put the synther's feature matching through its paces in comparison, so I also took a third photo at each location that rocked left or right pretty hard off balance, hence the title. I suppose to really test things I could have thrown in some photos that were upside down or on their side, but I would never do that intentionally in the first place in a regular synth, so I wasn't sure what the point was. Most photos, no matter how crooked, can be rotated in increments of 90 degrees to be at least as level as these.
Nathanael.Lawrence Over 1 year ago
I am curious to know what your thoughts are concerning the comparitave results.

Is there some big difference in the pointclouds that I'm missing? Did I not take the leveled set of photos like you had in mind? Should I try taking bigger sets of photos to see if a difference manifests with more input data?
Nathanael.Lawrence Over 1 year ago
My real question for you is, "Could it be true that pointcloud density (although you are clearly capable of picking the photos to generate it) is not tied to how level the shots are, as I understood you to be saying, but rather tied directly to how close you are to an object (greater fidelity descriptions of surfaces in pixels directly relates to number of unique surface markings that the algorithm can detect seems to be common sense to me), how visible its unique visual patterns are (is the subject a synthy object to begin with), and the number of different perspectives that cover any given area of the subject's visible surface, allowing the synther to track the comparitive movement of unique surface details in relation to each other to compute their positions in 3D space (as I understand the Photosynth Photography Guide's 'Rule of Three' [as a minimum] to be saying)?".
Nathanael.Lawrence Over 1 year ago
I feel as though we may be trying to solve different problems, as well. For you, I think the goal is simply, "I wish to create a dense pointcloud and pleasant navigation experience while spending the smallest amount of time/space on my memory card taking photos.".

This is not at all a bad goal and because the synther looks only at 1.5 or 2 megapixel versions of our photos, if we are providing photos of higher quality than 2 megapixels, it is possible to tease out more detail by cropping out smaller pieces of the original images and allowing the synther to inspect these pieces at 2 megapixels (or less, depending on the size of the crop), with the risk that the crops will artificially retain a point for every image feature found within their region of their parent photo's 2 megapixel version whose depth will be computed as a flat plane since there is no relative motion between surface details when comparing part of a photo to itself.
Nathanael.Lawrence Over 1 year ago
By contrast, my desire to orbit every major object in a scene does not satisfy your requirements of taking a minimum number of photos, however it is solving other problems which taking a minimal number of photos and then culling out more detail via cropping doesn't begin to solve.

My approach seeks to describe every object or part of an object such that, after synths can be linked or we see an option for people to register single photos of theirs against places which have been thouroughly synthed, there will be practically no angle which people can contribute a photo from which will be greater than 25 degrees different.

Whereas if a part of a building is symmetrical to the other side and taking overlapping sets of photos which only look straight at the wall leaves things ambiguous, my overlapping arcs leave nothing to chance, describing what is to the left and right of each portion of wall in a way that pivoting the camera in place cannot hope to compete with.
jimcseke Over 1 year ago
Hey Nathanael,
I still have some stuff to take care of, but I look forward to discussing your postings.
If you would rather do this in a more private manner, send me your email address or some other way to discuss this. If you would rather do it on this site, then I would be more than happy to as well.
I will be very frank and honest with my replies, so I hope that you will not take anything I say personally, as it will not be meant to be personal.
You are a good and decent person, but I think you have the ability to be very honest, and appreciate honesty in return.
The topic we are discussing goes beyond, this site and the photosynth website.
It involves more people than just you and I.
And I am going to be as frank and forthcoming with almost every detail which has led to this point.
There is more to this than meets the eye.
I am taking my very best non-human friend and companion (Mercedes my boxer buddy) for another hike.
jimcseke Over 1 year ago
When I return, it will only take me a short time to compose some of my responses.
I think that we will both find this to be very interesting.
A good informative chat between internet friends and anyone else watching, who might want to join in.
Above and beyond as usual Nathanael.
Thank you
Jim
Nathanael Over 1 year ago
I suppose that the problem which I am trying to solve could be stated this way:

"How can I synth an environment (a room, house, yard, cathedral, etc.) such that:
(#1) the surroundings are recognisable, even in pointcloud form,
(#2) any photo taken by any person for any reason within this space (wedding, birthday, holidays, etc. - completely unaware of Photosynth) stands a 90% or greater chance of synthing into my environment synth, even if furniture has been moved or plants have completely changed their soft tissue,
(#3) photos will align to both foreground *and* background objects, and
(#4) identical objects (such as two signs at the two entrances to a large parking lot, symmetrical architecture, etc.) are not confused with each other?".
Nathanael Over 1 year ago
Public or private makes no bones to me. My email address is nathanael.lawrence@live.com.

To quote a bit of pop culture, "Ask me the questions, Bridge Keeper; I am not afraid!".
jimcseke Over 1 year ago
Hey Nathanael,
I hope that i was able to address alot of your concerns, even though I know I was not very specific in regards to the side by sides which you provided. After I finished my post, I thought that perhaps there was a chance that I had not convinced you one way or another. And I know that, perhaps you wanted something that would at least hint to the possibility that my methods have merit. You mentioned before, that "talk is cheap" and I completely agree with you. But talk can also be valuable.
But I think in this case, the talk on my part may have been cheap. It was not intended to be. I love love the fact that you do not accept or reject until a certain degree of proof is provided.
Because your input has always been very valuable to me, I wanted especially to be convinced.
When I woke up this morning, a thought came to me.
jimcseke Over 1 year ago
Sorry I miss typed.
I wanted especially you to be convinced.
jimcseke Over 1 year ago
I thought of a way that might put you one step closer to at least see that near perfect forced camera orientation is essential for all of the reasons I have mentioned.
So I would like to ask your permission to take screen captures of the level versions, and attempt to edit them in a way, that in the very least I can prove to you that possibly, perhaps, and maybe my methods do have merit.You are a hard nut to crack and that is a very good trait.
I could possibly edit your "eyeballed" versions, but my editing does rely on a higher degree of certainty of camera orientation.
My LCD television's resolution won't provide a perfect replica of your images, but I think I can still use the lower resolution.
If you agree to this I will:
Increase the pointcloud density
Try to create higlights
Create unique images which provide a completely differen view of an original photograph.
None of these edit will be "artificial" by which I mean that everything I do will be part
jimcseke Over 1 year ago
of a fully functional synth.
As though all images I create were originals.
And may be judged in a 3d render program as to it's increased amount of usable data.
in other word I will multiply the number of photographs you originally captured as though you had captured the created image yourself.
You did say that you were a bit impatient and did not take as many photos as you would have wanted. This is another advantage to my methods.
During some of my initial experiments,I have tried to editing other public synths, but as I mentioned I neel a higher degree of cetainty as to the camera oriention. I don't know if this will be the last bit of proof you might need, but I think that it would be fairly close to the level of proof that you demand.
So I will wait for your okay and put my money where my mouth is.
I think it should be fun.
And when I am completed, I would like your opinion.
This is not the first time I have asked for your opinion.
jimcseke Over 1 year ago
Do you remember my train synth?
I asked you if you noticed any improvement due to some of the editing I had done on that particular synth.
You were very polite and honest, and mentioned that you liked the small pebbles by the track.
I respected your opinion then, and I respect your opinion now.
So let me know what you think, and then I can get started.
Dude, I am a human batch file editor. You have no idea how many images i have edited since I first started my testing.
I know that I have spent more time editing than I have taking the actual photographs.
I would have killed for a batch file editor that could do what I was doing, one photo at a time.
That was one of the main reasons that a batch file editor is part of my provisional patent.
There is no way that anyone in their right mind, would do what I did, just to create a better synth.
I am wondering how long it will be before i can look back at all of this and laugh. i'll give it about 2 years. :)
jimcseke Over 1 year ago
I will check back throughout the day and look for your response.
They are your photographs afterall.
Oh, by the way, I am also a hard nut to crack, so we have that in common.
Sincerely Jim
You need to be Signed In to add a comment.
New to Photosynth? Sign Up for a free account.