Visualsfm Manual Form: Fill & Download for Free

In Photosynth 1 (the original desktop application for Windows which is and has always been free) you are only constrained in how many photos you are able to match by how much RAM your computer has installed for the scene reconstruction phase.The largest photosynths that I have seen max out around 2,000 photos, however I don't know how many Photosynth users have the ideal 64-bit Windows computer with 8 GB or more of installed RAM.I, myself, have never had access to a computer with more than 4 GB of RAM installed and have been forced to divide some of my synths into smaller slices or less dense coverage but I have managed to put together upwards of 1,700 photos in a single synth with my limited setup.Getting the Most out of Photosynth32-BIT vs. 64-BIT Address SpacePhotosynth's desktop application (also referred to as the photosynther or 'synther' for short) is a 32-bit program and therefore can only use a maximum of 4 gigabytes of RAM at a time. To make full use of this capacity, you would use a 64-bit version of Windows (so that the operating system itself can use more that 4 gigabytes of RAM) on a computer with a 64-bit CPU and more than 4 gigabytes of RAM installed (6 or 8 gigabytes would be my minimum amount of RAM). This will allow Windows to run, using whatever RAM it needs, then leave a full 4 gigabytes of physical RAM free for Photosynth's process at all times.Note that you should probably not have much else open or processing on your computer while waiting for a synth to finish calculating, in order to protect the RAM that it needs.If you are using the synther on a 32-bit Windows computer with 4 gigabytes of RAM installed, then the maximum amount of physical RAM that each 32-bit application can use is 2 gigabytes (or 3 gigabytes if you've changed Windows' settings for the amount of RAM the OS gets vs. how much an application gets - a.k.a. the 3GB switch).Ideally, Microsoft would release a 64-bit version of the synther since 64-bit applications can use up to 8 terabytes of RAM (if you could afford a computer with that much RAM).Virtual MemoryIf you have a computer with less RAM than ideal, you can use Windows' 'Virtual Memory' paging file settings to specify a particular amount of your hard drive[s] to use for whatever data from open programs and files cannot fit in the available physical RAM. (I suggest the maximum 4 gigabytes size paging file on each hard drive on a 32-bit computer, provided that you have the free space available.) This will allow you to complete larger synths, however will not be as fast as if you had a 64-bit copy of Windows and more RAM since read/write times are slower on hard drives than they are on RAM. Also, your entire RAM will be filled as the operating system does its best to locate data in RAM which hasn't been accessed recently or won't be needed soon to shuffle over to the paging file and take whatever has been put in the paging file and put it back into physical RAM when needed which is not an efficient process.Storage SpaceIf you're using a computer with multiple hard drives, another thing to consider is using Windows' Environment Variables settings to make sure that Windows' TEMP folder is being stored on the hard drive with the most free space.The synther uses the TEMP folder to store all of its output files before they are uploaded to the Photosynth.net servers, so you'll need plenty of hard drive space to hold all of the converted JPGs you'll be uploading.(You can access your synth creation log files by typing %temp%/photosynther into any command line:Windows Explorer|My Computer address bar, the 'Run' command, etc.)To be on the safe side, you would highlight all of your input photos for a synth before starting, see how much space they take up on disk, then make sure that the hard drive where Windows' TEMP folder is stored has twice that much free space. Only after that would you open the synther and begin.Photo ResolutionThe resolution of the input photos is not critical for purposes of calculating the synth, since any photo larger than 1.5 megapixels will only have 1.5 megapixel copy of it examined for image features to match to other images.If you use photos with a lower resolution than 1.5 megapixels, you might be able to use a few more photos, however you're hurting Photosynth's vision at that point and also damaging the viewing experience later, since the full resolution photos are always uploaded for viewing. You said that you're only interested in the 3D reconstruction, but using input photos with lower resolution than 1.5 megapixels will result in a lower number of identified image features per image which will in turn mean fewer opportunities to match photos and fewer accurate points in the point cloud.Photo Concentration/DistributionAnother simple factor which will directly impact how many photos will fit in a synth before the synther runs out of RAM while reconstructing the point cloud and camera positions is how many of your photos see the same thing from close enough distances to create matches.Cement, concrete, rock, sand, rough tree bark, etc. are all things that are very easy for Photosynth to track.Large bodies of water, glossy reflective surfaces, distant masses of leaves blowing in the breeze, large smooth blank single-color walls, etc. are things that it doesn't track so well.How this applies:Scenario 1:I take a long stroll through a lakeside park where the trees are thinner on a breezy day, snapping photos as I simply walk forward. This may not generate much of a point cloud - in part because some parts of the view are not very synthy such as the surface of the lake - and in part because anything that is close to me does not appear in very many photos before it is out of view, and distant things do not create a very dense point cloud because they are so small in the context of a photo's frame.Scenario 2:I choose a single concrete statue or a shorter rough-barked tree with fat branches in that same park and take photos circling each major part of the statue or trunk/branches and get a massive point cloud with many fewer photos taken - in part because the material being photographed has a lot of speckled visual texture that is very easy for Photosynth to track - and also in part because all of the photos are very close to each other and see the same parts of the scene many times over from different points of view. This creates more matches and will require more RAM to hold all of the detected matches in memory while their positions are being found.Thus, it should be understood that the number of photos is not as important as how the photos which you have to use are spaced. In other words, having a tremendous amount of overlap will almost always result in superior reconstruction of a surface in detail, however you will certainly chew up your RAM faster with fewer images when doing so.Downloading and Viewing Photosynth DataThere are several options for downloading the point cloud and camera positions from a synth:Christoph Hausner's SynthExport presents you with a list of all point clouds in a synth to select and save in a variety of formats which can be viewed in 3D modeling applications like Meshlab. SynthExport can export the camera positions as a comma separated values file for viewing in Excel etc.Henri Astre's PhotoSynthToolkit was developed with the intent of getting Photosynth output (photos, camera positions, and sparse point cloud) to use as input to Yasutaka Furukawa's PMVS2 which had been designed to perform dense reconstruction using the sparse output of Noah Snavely's Bundler (the precursor to Microsoft's synther which Noah helped them build).To this end, PhotosynthToolkit downloads only the largest point cloud from a synth and its associated camera positions and was the first to offer the option to download the full resolution photos from an original synth.Although SynthExport now supports saving photos as well, Photosynthtoolkit remains the only one to preserve the input photos metadata (choose to download both the thumbs and HD copies of the photos, since the metadata is stored in the thumbs).MeshLab also features a built in plugin for importing data from an original photosynth, including the point cloud, camera positions, and photos. You do have to manually turn on per-vertex color and lighting to see the point cloud, though which is a bit user-unfriendly.Kean Walmsley of Autodesk wrote BrowsePhotosynth - a plugin for AutoCAD 2011 - to import the largest point cloud from a synth and manipulate in AutoCAD.Photogrammetry AlternativesAs far as photogrammetry alternatives to Photosynth, there are many.Look up Olafur Haraldsson's Photogrammetry Forum and Henri Astre's blog archives, both of which cover many of the advances in computer vision which were happening between 2010 and 2012.The Right Tool for the JobThe real question is, "What are you hoping to achieve?".Do you simply want to solve for as many camera positions as possible?If so, Photosynth won't be outdistanced by too many free options.Try a 64-bit version of Bundler or VisualSfM or (for pay) Photoscan.Are you primarily interested in a point cloud?Would a sparse reconstruction like Photosynth's or Bundler's be enough for you or do you want a dense reconstruction like PMVS2 provides afterward?Are you hoping for a textured mesh as output? In that case, you may want to look at Autodesk's 123D Catch or (if you're willing to pay) Agisoft's PhotoScan.Several of the webservice options listed at the Photogrammetry forum are now offline (My3DScanner, Hypr3D, PhotoCity) but even when they were online were often constrained in the number of photos able to be accepted as input (due to the storage and computational limits of free hosted projects run by individuals or small teams, rather than large companies like Microsoft or AutoDesk).Your best bet for sheer number of images able to be matched is something which uses your own computer to do the computer vision calculations like Photosynth's desktop application, or a 64-bit version of Bundler, or Agisoft's Photoscan.A couple of years ago impressive advances in research were made in speeding up reconstruction and handling more images by utilizing GPGPU computing and one free example is available in the form of ChangChang Wu's VisualSfM.Photosynth 2 and BeyondOn 2013 December 10, the Photosynth team opened access to the Photosynth 2 beta (a.k.a. Photosynth 2014 Technical Preview). This version of their technology is a web service where any Photosynth account can upload a maximum of 200 photos per synth. The synth is computed on Photosynth's servers in Azure and then displayed in any modern web browser using WebGL for 3D graphics on the user's computer.Unlike Photosynth 1, the point clouds are hidden and no attempt is made to extend the view past the input frame.A smoothed virtual camera path is also produced which travels as closely to the solved for camera positions as possible while maintaining a smooth curve or straight line.A major distinguishing feature from original photosynths is that Photosynth 2 produces a low-polygon-count depth map for each photo which the photo is then projected onto and every two photos crossfade between their pixels and depth maps as the virtual camera moves through the imagery.Since Photosynth 2 synths are designed to travel only one direction in no more than 200 photos, the ambiguity of where a user might choose to explore is diminished and thankfully the Photosynth team has taken this opportunity to load a medium resolution copy of each photo in the synth as quickly as possible in advance of your arrival at it which makes keeping your eye on an object while moving from photo to photo far superior to Photosynth 1's philosophy of 'begin loading the image only after the user has arrived at it'.Operating at Video ScaleYou may be wondering why I am mentioning Photosynth 2 if it puts together fewer photos than the desktop application. It is because a recent Microsoft Research project (First-person Hyperlapse Videos) has taken the Photosynth 2 ideas and extended them to long videos which, as you can imagine, are many more fames in length than either Photosynth 2's 200 photos or Photosynth 1's 2,000 photos.It is true that the Hyperlapse project still presents fewer output frames than were input, however it does so, based on the solved for camera motion from (I believe) all of the input video frames.