Sparse reconstruction

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • mr_anderson
    3Dflower
    • Aug 2019
    • 4

    Sparse reconstruction

    Hello, I am quite new to photogrammetry and am really enjoying your product.

    If after using fast preset to create sparse point cloud all of my cameras are correctly oriented, can I get better results when using slower, better settings or it doesn't matter?
  • Roberto
    3Dflow
    • Jun 2011
    • 559

    #2
    Hi Mr. Anderson,

    Usually, default and deep presets should work better than the fast preset. However, the fast preset has a lower keypoints sensitivity, and it will extract fewer key points. While it might sound counterintuitive, less is better in some cases. For example, when you have micro noise in pictures at the pixel level, or you are in turntable scenario with a uniform background.

    Comment

    • mr_anderson
      3Dflower
      • Aug 2019
      • 4

      #3
      So if I wanted to make the best out of the pictures I have, and time is not a factor, deep will work better than default in most cases? Or are you saying it is possible to make things worse by changing default to deep?
      Also what do you mean by micro noise at pixel level?

      Thanks for you reply

      Comment

      • cam3d
        3Dflover
        • Sep 2017
        • 661

        #4
        mr_anderson It's a case by case sort of deal, heavily depends on the data-set. What kind of things are you thinking about processing? - (Ps. It's definitely possible to make things worse by changing default to deep - highly not recommended. If you can acquire a better data-set it's almost always preferred to go that route rather than fiddling around with processing settings).

        Micro noise at pixel level would be sensor noise I assume - For example if you're shooting at a high ISO like 1600+ (depending on camera) there will be a lot more random noise variation than with ISO 100 on the same camera.

        Comment

        • Pappa Mark
          3Dflower
          • Aug 2019
          • 4

          #5
          Am trying to process the free pose photo sets from posespace.com

          they look ideal with 24 shots at 15 degree angles

          all i get w 3df zephyr i a few sparse points

          am I missing something in my settings ?

          Comment

          • cam3d
            3Dflover
            • Sep 2017
            • 661

            #6
            Hi Pappa Mark, without looking at the specific data, my thoughts are that 24 shots at 15 degree angles isn't enough to reconstruct from - If you can share the dataset I can give more feedback.

            Comment

            • Pappa Mark
              3Dflower
              • Aug 2019
              • 4

              #7
              love to
              lets see if this works
              did all 24 make it or just three?

              Pappa Mark

              Comment

              • Andrea Alessi
                3Dflow Staff
                • Oct 2013
                • 1304

                #8
                Hi Pappa Mark,

                I had to remove your images as they were 1. copyrighted and 2. showing nudity - sorry, I'm sure you understand!

                Regardless, it's better to share data with us with either a public link (e.g. dropbox link) or by using a third paty service (e.g. wetransfer) to support@3dflow.net

                As for your dataset:

                - the images are very low-res (566x900) - while you can use them, you won't be able to expect great results with such a low number of photos at that resolution (unless you are VERY close to the subject).
                - this seems to be a turntable setup, so you mask out the background
                - you would probably need more overlap between each shot, and a more appropriate lighting (less shadows) - although probably with a higher resolution the lighting is the last of your problem in this case
                - it's hard to tell from this dataset as it is low res, but is it an actor or a statue? If it's a real human being, making it stand while rotating around him / have it on a rotating platform, will certainly cause the subject to move - that's the most critical thing - the subject must be completely still. Either make a very quick capture (using a video may help) or use synchronized camera rigs for this purpose

                Comment

                • Pappa Mark
                  3Dflower
                  • Aug 2019
                  • 4

                  #9
                  ok

                  set had a user agreement that allowed for use in art statues as long as not used as is and had to be modified

                  they had a similar clothed set but it was huge so i sent the one that fit

                  did not know photos violated policy . I apologize

                  I did try masking to no avail.

                  the way the shadows move my guess is they use four cameras in a fixed room and then pivot model somehow.

                  Moving on - Kudos for a great product

                  I find the free Zephyr software excellent in my normal use of a 50 shot step count click circle around my subject with a simple iphone6 during twilight or a cloudy day.
                  The ply and obj exports usually deliver large blocky items (files too large ? )but the Glb export is clean and I return it to stl format with 3D Builder.

                  Thank you for the reply and I hope my pipeline helps other beginners

                  Pappa Mark



                  Comment

                  • cam3d
                    3Dflover
                    • Sep 2017
                    • 661

                    #10
                    Pappa Mark - Large blocky items? Can you provide a picture of this issue so we can help figure out what the problem is?

                    Comment

                    Working...
                    😀
                    😂
                    🥰
                    😘
                    🤢
                    😎
                    😞
                    😡
                    👍
                    👎