The ability to selectively add or delete photos to a Photosynth.

  • 27
  • Idea
  • Updated 8 years ago
This way I can remove photos that don't work or are redundant, and add additional photos to make the scene more synthy.
Photo of jtrpop

jtrpop

  • 7 Posts
  • 0 Reply Likes
  • making my Photosynths more synthy!

Posted 12 years ago

  • 27
Photo of Jonathan

Jonathan, Official Rep

  • 98 Posts
  • 57 Reply Likes
This is something we want to add, too!
Photo of stewATPBw

stewATPBw

  • 1 Post
  • 0 Reply Likes
Absolutely necessary. After it creates the synth I'd like to be able to review it using the viewer and right-click to remove images, especially those that only have 1-2 images in the 3D group (as in the control Switch to Next 3D group).

Thanks!
Photo of MickY_G

MickY_G

  • 1 Post
  • 0 Reply Likes
I'm pissed off that typically of MS, they didn't think to include the obvious in this software - as usual!
Photo of Nathanael Lawrence

Nathanael Lawrence

  • 795 Posts
  • 55 Reply Likes
MickY_G, I am somewhat amazed at you. Do you really believe that Live Labs didn't think of this? As Jonathan stated above, they want and plan to implement this feature.

What we have today is simply the very first version. Live Labs got the synther to a point where it could successfully run on almost any Windows PC at an amazing speed compared to the original tests (hours instead of days) and wanted to let all of us who had been dying to try synthing our own photos have a go at it. They aren't finished. They have plenty of things that they want to add. Just go watch some of their videos around the internet to see that they have much bigger dreams than this first release.

We are getting to play with Photosynth when the features are not even all added yet. We are lucky! Imagine if they had made us wait one more year until they add all sorts of refinements and features like the one asked for here. I'm just glad we get to use it now. It will get much much better.
Photo of bitplane

bitplane

  • 56 Posts
  • 11 Reply Likes
When you add this idea, it would also be cool to be able to hide/remove images of a certain size.
This would allow us to use low quality video frames to build up the features and the point cloud, and then stitch some selected high quality images into the 3D scene. I've made some 1300+ image synths from streams of LQ images and they work much quicker than a 300 image HQ one.

I suppose the big UI problem is how do you navigate between images if they no longer overlap?
Photo of Nathanael Lawrence

Nathanael Lawrence

  • 795 Posts
  • 55 Reply Likes
The space bar, zed, comma, and full stop keys would still accomplish navigation quite well. As you point out, it's not quite the neighbour tour if you've suppressed overlapping images, but it would get the job done. :)
Photo of markp

markp

  • 47 Posts
  • 3 Reply Likes
Problem is that doing something like that would take an enormous amount of processing power itself - likely several times more than the average "per picture" time as it then has to check against all the existing images in the synth, make new paths relating it to them (and them back to it), etc. EG for a 200 picture one that takes 3 hours to process, easily more than a minute per picture to stitch in a new one - probably 5-10 mins or more. Not something you really want to be doing via a silverlight client over a consumer broadband connection, nor having to bodge it and offload the strain to the servers. Similarly there'd be extra work to patch the hole left by deleting a broken one, unless it genuinely added nothing to the synth (and even then, there'd have to be recalculation if the the suspicion is that it was disrupting the synthiness of everything else somehow).

This, I suspect, is why it hasn't been implemented and we end up having to do over from scratch .... because inserting or removing (or both), say, 10 images, would take almost as long as simply doing a whole new synth AND yield less satisfactory results.
Photo of Mat

Mat

  • 1 Post
  • 0 Reply Likes
It's not our problem !
Photo of markp

markp

  • 47 Posts
  • 3 Reply Likes
Actually I'd like to retract some of that after re-reading it and the rest of the thread and thinking a bit further, given....
a/ If you approached it from the angle that these pictures might have been the last in the set anyway, there's no reason they couldn't be batch-added (no interactivity, just tell your client software "put these in there" and let it pull all the data off the MS servers, process it in the background then send back the changed parts), and conceptually the only extra processing would be that which those "last pictures" would have incurred anyway, had they been part of the set to begin with.
b/ deleting pictures would be a fairly low-cost exercise - all the others could retain their already-calculated positions and relationships, and only the transitions/navigation/display order would have to be patched. There could be an option on whether or not to delete that image's cloud points also - as you may be deleting it because of getting someone's head in the frame who doesn't want to appear online, or in other cases because it's stitched into the synth in an entirely wrong way, corrupting everything else. (With, I suppose, a third option of "fully recalculate synth"... which you could argue isn't much better than starting over from scratch but has the benefits that 1. all the other images have already been resized and uploaded, 2. you might not have those files on your PC or otherwise accessible any more, 3. there's scope for extra code that can more rapidly determine whether or not the missing image actually has any bearing on various other groups (e.g. if there's a set of pics "round the corner" from a removed shot, its removal is completely irrelevant) and cut down on calculation time)

And re: "not our problem"... well, it sort of is. No matter how neat the suggestion is, if the practical considerations can't be met, then it won't happen. In this case, if it becomes detrimental to the service that the company is already providing to most other users of PS (or to the performance of the MS servers as a whole) without providing a great deal of net benefit, then they likely won't do it.
Photo of robofill

robofill

  • 2 Posts
  • 0 Reply Likes
Well im sure their working on it.
For example i have a lot of Forty-Niners pics, one of them includes a cartoon leaking on the Raider logo. After this was Sythed it turns out its leaking on the head of my own Chearleader team..not good, it has to be edited or ask at least if we like it the way it is or re-organize the fotos in diffrent locations.
Photo of Nathanael Lawrence

Nathanael Lawrence

  • 795 Posts
  • 55 Reply Likes
Robofill, I've meant to comment on this ever since you posted it.

Since Photosynth uses Seadragon for the image presentation, the layout of the images in Grid Mode is completely dynamic, meaning that if you switch to Grid Mode while your browser window is maximised, you'll get a different layout than if you change your browser to half of your screen and switch to Grid Mode from there. The same idea applies if you have a widescreen monitor and I have a traditional 4:3 tube monitor.

Because of this, the number of people who see your synth as you do with Calvin letting loose on the 49ers girls is probably not as high as you think.
Photo of Erik

Erik

  • 1 Post
  • 0 Reply Likes
One year on and no ability to add, or delete photos without redoing a full synth, and making paper and pencil notes on what to include the next time.

Makes me think this is just a lab experiment that has no future in MS eyes. Don't count on your synths being around as long as you would like.
Photo of Nathanael Lawrence

Nathanael Lawrence

  • 795 Posts
  • 55 Reply Likes
I meant to reply to this back when you first wrote it, Erik.

Let's consider some facts.

1. You don't need to make paper and pencil notes on what to include next time. Just open your %temp%\photosynther folder and read your log file. There will be one for each synth that you have completed on the current computer and it will include a list of exactly which images stuck together in each group in the synth.

2. Photosynth left Live Labs some time ago (not long at all after they released it last year, actually) and is now part of the Virtual Earth or Bing Maps group.

3. Photosynths were just integrated into the beta Silverlight version of Bing Maps.

4. Computer Vision is a new and growing field that is only going to get bigger in the computer industry from now on.

5. Google, Adobe, and many others will not stop their Computer Vision research or products.

6. Microsoft is currently one of the companies currently at the head of the pack in Computer Vision. Photosynth is a perfect way for them to test their newest CV algorithms out on thousands of people's photos and costs Microsoft no money to do so. The only alternative to keeping this project and related research alive is to forfeit to Google, etc.

7. Just as in Google Maps, there are many places in Bing Maps without current imagery. One of the questions that Bing Maps employees hear all the time is, "When will you add imagery of my house/capitol building/house of worship/town to the map?".

8. Photosynth gives people a way to put whatever they have rights to photograph on the map for themselves (instead of waiting for Microsoft's vans to drive by, planes to fly over, or satellites that they have hired to orbit overhead). This turns Bing Maps from something like Encarta into something like Wikipedia or from something like Blockbuster Video into something like YouTube or Vimeo.

9. Photosynth is one of the things that even people who hate Microsoft admit is pretty great.

Now you tell me, is Photosynth likely to go away anytime soon? No; I don't think so.

I've listened to Noah Snavely (the guy who first implemented Photo Tourism in his Ph.D. work at the University of Washington before he graduated) and from what he says, it is relatively easy to add a photo to a collection once the initial large group is calculated.

I suspect that what has kept the Photosynth team from implementing this are the following:


  • Photosynth was designed from the beginning so that there would be one central place for synths to live so that one day they can start linking together. Because of this, they are stored online only.


  • As Tahrey points out above, the process of synthing intrinsically means comparing one photo to every other photo in the collection.


  • If all the photos in a particular synth are now on the server and the photos you want to add are on your computer, that either means that: (1) the Photosynth servers would have to do the synthing or (2) that there would have to be some way for the synther to download both the original synth metadata files (the photo positions and pointcloud, etc.) as well as small versions of all the images in the online synth to compare to whatever new photos you're adding. (Depending on the original photos still being the same place on your hard drive, named the same, and/or unedited since you uploaded them the first time is too unreliable.)


  • Because synthing is an extremely taxing computation, the Photosynth team has stayed away from doing any real computation of synths on their servers. Maybe they will someday if they introduce mobile synthing, but there are no real promises of that. I would imagine that they would wait to do that until they have a decent sized database of imagery for a fair number of towns. It might be other Photosynths in the area or just coverage with their Streetside imagery.


  • As things currently stand, each synth is perfectly preserved from the day that it was made. If the Photosynth team has come up with a slightly different method for the synther to use to try to make some synths work that didn't before but seem like they should, they have the original synth to compare new test results to. Simply copy the images from any troubled synth and run them in the updated experimental synther back at Photosynth HQ and then compare the two results. If they open up editing then you mix up all the results because people will be using new versions of the synther to edit older synths, etc. etc.


That last point is not really any good reason in my mind to keep from opening up editing, as synth metadata files are not really all that big and they could make something of a wiki approach for the synth database, where all previous versions of a synth are archived exactly as they were but new edits can always be added.

To wrap up on the topic of why adding, updating, or removing photos hasn't been introduced yet, these are just my best guesses. The Photosynth team could tell you what the real reasons are and I'm sure that there are many far more technical reasons than I have even thought about, but maybe some of my hunches are correct and they should get you thinking in the right direction anyway. If it's any consolation to you, quite a few of the members of the Photosynth team, themselves, have posted around various forums that they really wish that editing synths was already in place, so you're not alone in your frustrations. Their team is simply not massive and has many more features to implement than they have people to build said features. Other things like getting Photosynths onto Bing Maps were simply higher on the list than adding or removing photos.

As a closing shot, I'll just point out that even if we had a synther that could download the appropriate data from synths that are currently online in your account and add more photos to them, you'd still be waiting for each new photo to be compared to all existing photos and you'd still be waiting on only the new photos to be uploaded (even from the first day the synther didn't upload the same photos twice unless you'd edited them since the first upload). This makes me ask the question, "If editing a synth would look exactly like making a new version of a synth and take the same amount of time, except for adding more time to download existing synth data, then why are we unhappy with just making a new version of the synth using our existing photos from previous versions plus the new ones?". The difference, I suppose, is that if in downloading the current synth metadata, you could simply continue the calculation from that point, rather than needing to redo all of it, but adding a single image could easily change the computed position of many other images. For an example of what I mean, download this zip file of the Photo Tourism presentation (originally found on this page and open the video TreviReconstruct2b.wmv.
Photo of markp

markp

  • 47 Posts
  • 3 Reply Likes
Nathaniel... something I'd been meaning to try recently but haven't been able to get the time to play with it (only here because reply notification landed in my email ;) was sneakily taking a load of screenshot images from, e.g. Google or Bing street view and seeing if some kind of synth could be made out of them, whether the engine would accept them and come up with some kind of mindblowingly cool 3D translation of those images. It'd be locked to the path the car took of course, and all the buildings would be at a single distance from the cam (so something completely along the lines of the classic Venice synth may not work, unless we could also use an auto-stitch feature to create panoramas/extended width shots as part of the process), but would create an excellent base to build upon, for 3D models, integrating 3rd party and user synths & images, etc.

I haven't even had chance to properly try out Bing streetview and the synth integration with that yet, so maybe it even works on that level already :D --- would something that works that way be a feature you'd look to include in future? Go straight and near-seamlessly from the "official", baseline streetview made in a rather crude fashion by someone driving a car and getting time-lapse imagery, into a more detailed synth submitted by joe random focussing on a particular point of interest. EG for one thing in my own city, the camera car might have driven past Spitfire Island (a roundabout/traffic circle with a sculpture on it commemorating an important WW2 airfield, and factory making the eponymous aircraft flying from it) without paying it any special attention (it's partway along a major road, after all), but there'd be zoom panels taking you to one - or more, spliced? - synths of the actual sculpture made by a local shutterbug. And maybe even the same in reverse...?
Photo of markp

markp

  • 47 Posts
  • 3 Reply Likes
Nathaniel... something I'd been meaning to try recently but haven't been able to get the time to play with it (only here because reply notification landed in my email ;) was sneakily taking a load of screenshot images from, e.g. Google or Bing street view and seeing if some kind of synth could be made out of them, whether the engine would accept them and come up with some kind of mindblowingly cool 3D translation of those images. It'd be locked to the path the car took of course, and all the buildings would be at a single distance from the cam (so something completely along the lines of the classic Venice synth may not work, unless we could also use an auto-stitch feature to create panoramas/extended width shots as part of the process), but would create an excellent base to build upon, for 3D models, integrating 3rd party and user synths & images, etc.

I haven't even had chance to properly try out Bing streetview and the synth integration with that yet, so maybe it even works on that level already :D --- would something that works that way be a feature you'd look to include in future? Go straight and near-seamlessly from the "official", baseline streetview made in a rather crude fashion by someone driving a car and getting time-lapse imagery, into a more detailed synth submitted by joe random focussing on a particular point of interest. EG for one thing in my own city, the camera car might have driven past Spitfire Island (a roundabout/traffic circle with a sculpture on it commemorating an important WW2 airfield, and factory making the eponymous aircraft flying from it) without paying it any special attention (it's partway along a major road, after all), but there'd be zoom panels taking you to one - or more, spliced? - synths of the actual sculpture made by a local shutterbug. And maybe even the same in reverse...?
Photo of Stefan

Stefan

  • 1 Post
  • 0 Reply Likes
Lets hope this comes out soon, I just did a test, and it did not work very well, want to delete the whole thing and start over again :P
Photo of Ryan D Kyle

Ryan D Kyle

  • 1 Post
  • 2 Reply Likes
Has any progress been made on this feature request? No comments in 3 years...

It would be super helpful. My scenario is that I've built a couple synths, one with nearly 200 images, and I'd like to remove just an image or two.

Thanks for the consideration of this feature. Please put it on the 'take a look at again' list, please.