Reconstruction config comparison
1 year, 7 months agoI’ve made a video showing the results of dense point-cloud reconstruction with various settings on the sample ET imageset for Bundler. This should give a quick visual reference of what to expect with different settings.
Subscribe to all content
Very nice video Josh and indeed proving some non-linear results. I have a question on your Bundler Photogrammetry Package. When I try to use it – even on heavily downsampled images – I often get the warning: too many features – reallocate textures. Does it mean my images are still too big? So should I increase My RAM and swap file or change the size of my images?
Sometimes, but not always, there is no *. ply output in the bundle folder. Is this due to this message or something else. Thanks for the help and keep up the good work!
All the best, gEEvEE
@gEEvEE: The lack of .ply files is usually just because no good matches have been found. I have had good results out of imagesets that occasionally give the “too many features” error, though you might want to change the last parameter in the sparseRecon64.bat file to 1, i.e. BundlerMatcher list.txt gpu.matches.txt 0.8 1
Thanks for the info Josh, but it was not a great success either. It is strange, I have a bunch of around 40 photographs that are very nicely synthed by Photosynth or Photoscan (another terrific piece of software form Agisoft), but whatever I try with the Bundler Photogrammetry Package, I never get more than 2 ply files (and in most occasions none). The ET example works very fine, but there Photosynth can not create a 100 % synth. So has it something to do with the images or the apparent structure in them? I can always upload you some data if you want to test it yourself.
I am currently comparing all the packages for a paper, so it would be really nice if you could help me further with this.
All the best, gEEvEE
@gEEvEE: Sure thing – feel free to send your imageset and I’ll have a go at reconstructing it. My email address is josh (dot) harle (at) gmail (dot) com
One significant difference between PhotoSynth and Bundler is how dependent they are on given camera properties. Have you added your cameras CCD size to the extract_focal.pl file list? This may require a little internet research to find, but can make a big difference. Also, PhotoSynth does match features differently to Bundler. I’ve noticed that PhotoSynth will consistently get more matches, but occasionally (at least for me) some of these matches are false.
Thanks! I wrote you an email with some remarks and questions. Cheers
Sometimes i have the gEEvEE’s problem, no .ply output. I solved in my last test modifying the sparseRecon64.bat.
I realized that with less “images per segment” i get more ply output and a denser point cloud.
Which are the optimal photocamera settings (image size,flash and so on)? Thanks.
@Micro;
Ideally use the highest image size you can capture initially, and then resize if there are too many points. For shot illumination, try not to use direct flash: you want an indirect light that will not cause shifting highlights or reflections on your subject. The PhotoSynth Photography Guide is useful and general guide that applies to any photogrammetry project.
@josh
Which software do you suggest to resize images? Any particular parameters in resizing?
After the resizing do i need to create exif of every images?
how are the results of pmvs after a resizing of images?
Thanks a lot.
@marco,
I use PhotoSynth, as I can easily do batch resize with the “Actions” (although you can do this with ImageMagick too, it’s harder to set up).
Yes you will need to preserve the EXIF data when you resize, so use “save” rather than “export” using PhotoShop, or just check in the file details after any other method to confirm they have not been removed.