Forum : New Feature Suggestions/Requests

Do you have an idea for an awesome feature we should add… or hate the way we’re currently doing something? Share your ideas and suggestions here.


Topic: manual synth

Report Abuse Report Abuse
Carl_Lin (Over 1 year ago)
I love this software, but I have one suggestion. It seems sometime the computer can't synth two objects in the photoes together even though they are the same, so I think it will be great if there is a funciton which lets me partially overlap two pictures together manually and let the computer synth it.
dariusmonsef (Over 1 year ago)
This is something that has been requested before and it would be a useful feature... we just need to figure out an easy way to implement the feature along with the many other great suggestions we've heard.  Thanks for the feedback.
Andrew0002 (Over 1 year ago)
Yeah. This should be a duh thing. I have synths that only get 30%, but there are obvious relations between 3D groups that just don't get put together somehow.
Nathanael (Over 1 year ago)
Andrew, I'm with you, but if your subject is something that you can take more photos of, you could always try taking more photos in between the spots that didn't connect and try making a new version of the synth with the new and old combined.

One other trick to try if you're synthing something that you can't take more photos of is to make copies of your long shot photos and crop out the things that should be connecting to close-ups. Because Photosynth uses smaller versions of your photos during the photo matching sequence (so it doesn't find so many details that it either can't finish putting them all together, also so it can put together more photos) sometimes details of far off objects get overlooked by the image matching because they are so small in the versions of your photos that get matched to each other. This simulates taking photos between close ups and wide shots but really is not as good as being able to actually take more photos between the two locations
macj (Over 1 year ago)
Amplifying on hte above suggestions, why not a totally manual version for power users, where a simple org-chart function is used to manually lay out all of the connections we wish to be Synthed.  Power users are control freaks, and we don't want no freakin' software to tell us what photos go together.  Plus, many users have libraries of photos not shot for synth, but which would be great if manually put together.
Nathanael.Lawrence (Over 1 year ago)
A manual override would be much appreciated by me, but really only as a last resort. When dealing with a few hundred images, I definitely want Photosynth to match as much as it can on its own.

My ideal is more along the lines of (after Photosynth takes a pass at matching the group) I am able to: 
(1) point out connections that it didn't make at all & 
(2) strengthen weak (a.k.a. loose/floppy/portal) connections by adding more reference points to better fit oddly angled photos into place.
After my input, Photosynth could reprocess the portions of the images that I have highlighted for comparison with the full resolution versions of those photos (so as to find more image features to match with the hopes of getting it automatically this time around).

If all of that fails, though, please give me manual override. I would love to see a sort of accuracy voting system built into Photosynth that allows me to inform Photosynth statistically what end users accept as 'good'.
Nathanael.Lawrence (Over 1 year ago)
To deal with spammers, a ranking system could be introduced to recognise users whose judgements on images' placement are consistently 'seconded' by others (keeping a watchful eye on whatever the common loopholes of such systems are, i.e. these members all vote each other up all the time - are they our most helpful users or are they just spammers looking to up their reputation rank?). 

By establishing reputation, the system could take the feedback from trustworthy users more weightily. I'm not sure how this would be used to affect the algorithms, but surely some data would surface up - common variables where success or failure is consistently present. I recall Polar Rose using a similar feedback system for their facial recognition plugin, but their case is substantially easier (to say the least) than the complexity that is a synth.

If such a feedback system is infeasible, I would be interested in hearing why that is.
You need to be Signed In to add a comment. (Are you new? Sign Up for a free account.)