Topic: exact same photos from months before, but inferior results.

Report Abuse Report Abuse
mzaloon (Over 1 year ago)
Hi Photosynth developers!

I want to report a strange thing on current photosynth algorithms..

i make the exact same synth, from exactly same collection that i was months before, and the current results of synth, are poor that my initial synth. however the new poor jumps from 31% to 41% synthy.

here, are the links to the synths.

the synth is basically a very old (haunted?) 3 floor house. 1229 photos.

NOTE: the 31% synthed are better, with no error co-relations between photos.
the 41% have a some error co-relating the photos. (the collection of photos is exacly the same!)

poor 41%

better 31%

thanks guys! I will wait for some feedback! ;)
Malcolm -
Nathanael (Over 1 year ago)
Hey, Malcolm,

Here's two topics that have gotten some replies from the Photosynth team that address the same topic that might interest you:

I've bumped into this quite a few times now, when synthing over 1000 photos. It's not as much a result of the number of photos, I don't think, but rather a failure to provide good clearly defined groups and paths between those groups.

As far as the percentages on the synths that you listed, the percentage is primarily influenced by how many of the photos tie to each other. If Photosynth makes false matches, then it may well mean that more photos are linked but you may end up with wrongly aligned groups in the pointcloud. That makes for a higher percentage but lower quality as you say.
Nathanael (Over 1 year ago)
Anyhow, I'm glad you asked. I just thought you might like to see a few answers in case no new ones appear here anytime soon.

One last thought is that depending on how long it's been between the two attempts, the synther may have been updated to work slightly differently since the last time. In the end it's a roll of the dice as to whether the synth you see is the best that can possibly be, given that set of photos.
madeeds (Over 1 year ago)
Nathanael really hit the nail on the head with his reply.  One of these was created in October, and one in February of this year.  The more recent one got more matches from the synther, but as you pointed out, they are not great matches, and this causes the point cloud to be larger, but worse.  It looks like there are at least three different coordinate systems that are joined poorly with the later synth.

Great subject matter by the way.  The synther loves this sort of thing.  To get this to be 100% synthy you probably need to be a little more robotic in the way you take your photos.  One or two complete panos per room, shots taken every 3 feet (coming and going) works pretty well.

Here's a great example:
mzaloon (Over 1 year ago)
Hi Madeeds!

i´m thinking, the order of add photos, matter? this is only thing i can changed for mistake on this comparative.

i will make another synth with very different order of add photos... i post it soon...
and i pretend to back to this hunted house, to take more a hundred of photos... ;)

are you developer?
thanks again! ;)
mzaloon (Over 1 year ago)
here go...
same collection, diferent order.. much improved result over second try.
this time, 45% synthy, with no wrong co-relations!

i use Softimage ICE, do you know it?!
I really like to use this pointclouds into ICE, it will be possible? (i saw the web "ripper" thing, but never works on my tries)

Nathanael (Over 1 year ago)
Congratulations on the improvement! It does really seem wise to set your photos in the synther in a logical order if your first attempt with a collection doesn't come out how you like or expected.

As to exporting your pointcloud and using it in Softimage ICE... I'll have to get back to you on that. Finding an answer requires more time than I've got at the moment.

Also, yes, madeeds is a Photosynth developer.
mzaloon (Over 1 year ago)
sorry for the double post...
the Softimage ICE  Photosynth thing, i wait for your words.

thanks Nathanael!