Forum

Author Topic: Extreme distortion of the sparse cloud using masks?  (Read 3397 times)

jerry7171

  • Newbie
  • *
  • Posts: 27
    • View Profile
    • Thingiverse Profile
Extreme distortion of the sparse cloud using masks?
« on: April 18, 2016, 11:25:20 PM »
Hello,

I've been obsessing over a full-sized bronze copy of Michelangelo's David that stands here in my hometown. I've had mixed results at best but recently tried again.

I took over 500+ photos with a simple point & shoot Canon, imported the photos into Lightroom and set to work correcting the images for exposure, etc. I know that goes against the advice given in the manual, but with the subject being a dark, shiny bronze nude male, I thought it couldn't hurt to try a different approach.

I opened the folder containing all the photos in Photoscan and put it to work using the "Low" setting for each step.

You can imagine my surprise and joy when I came back hours later to discover that Photoscan not only used every photo but created a very good model. I only had to adjust the box around the statue, remove the fringes of overcast sky that had been piled like Marie Antionette's hair atop the statue's head and set Photoscan to work creating a dense cloud and finally a mesh.

I used the final, low-quality setting mesh model to create masks so I could turn around and start over with "High" settings and speed the process along.

It worked, but now I end up with some really wacky results that I can't figure out as shown by the screenshots below. What happened? It seems almost like using masks and focusing Photoscan on the statue made things worse.

I'm using a 2009 Mac Pro, 2.66GHz quad-core Intel Xeon with 32GB of RAM, NVIDIA GeForce GTX 960 with 4GB of onboard RAM.

stihl

  • Sr. Member
  • ****
  • Posts: 410
    • View Profile
Re: Extreme distortion of the sparse cloud using masks?
« Reply #1 on: April 18, 2016, 11:37:49 PM »
Not sure what you're referring to. Do you mean that the statue isn't attached/aligned to the base of the statue?
As for a sparce cloud this looks fine to me. Needs some gradual selection and manual editing to tidy it up. Trying using the Image count in gradual selection and put it on value 2. The rest can be dealt with by deleting the outliers manually.

jerry7171

  • Newbie
  • *
  • Posts: 27
    • View Profile
    • Thingiverse Profile
Re: Extreme distortion of the sparse cloud using masks?
« Reply #2 on: April 19, 2016, 12:05:06 AM »
Hi Stihl,

I guess I was perplexed by the outcome differences from the first time to the second time. Without any real input from me, Photoscan at the "Low" setting was able to create a sparse cloud with no distortions of the statue in the first try. On the second try, suddenly the sparse cloud acted like string cheese in some areas. That was what threw me off. I thought that using masks was a nearly ironclad way of guaranteeing no problems or need for extensive cleanup, plus faster model creation?

After I posted here, I started looking more closely at the sparse cloud model and began editing it, just as you suggested. I think I've cleaned up it enough to see where it will go.

I can't help but still be impressed that Photoscan was able to do all this with the photos from a low end Canon point & shoot that only outputs 14 megapixel JPGs. I usually prefer to use my Pentax in RAW mode, but for some reason I had less success with this statue in those attempts.

sc

  • Jr. Member
  • **
  • Posts: 59
    • View Profile
Re: Extreme distortion of the sparse cloud using masks?
« Reply #3 on: April 19, 2016, 11:15:12 AM »
you can try creating masks from the model you get from the sparse cloud or a dense cloud on low settings, then redo the alignment with the new masks

FLuca

  • Jr. Member
  • **
  • Posts: 57
    • View Profile
    • Another-me
Re: Extreme distortion of the sparse cloud using masks?
« Reply #4 on: April 19, 2016, 03:43:09 PM »
Hi Jerry,
It would be nice to have a picture of the statue to confirm my words but I suspect that the statue is shinny or specular. In this case, the camera alignement has been made properly due to the environment points.
When you have restricted the matching points to the statue with your masks, there where not enough matching or many wrong matching resulting in bad camera alignment.
In your specific case, I think it is preferable to keep the environment for alignment, or limit the masked area to unfocused areas, such as out of DOF, or changed areas such as people moving around.
Good luck and send screenshot of you final model

Fred Lucazeau - Body Scan - www.another-me.fr
Sketchfab gallery: https://sketchfab.com/fredlucazeau

jerry7171

  • Newbie
  • *
  • Posts: 27
    • View Profile
    • Thingiverse Profile
Re: Extreme distortion of the sparse cloud using masks?
« Reply #5 on: April 19, 2016, 10:49:26 PM »
Here are some screenshots of what I'm working with...

I've included a couple of Adobe Lightroom screenshots that show the 'Before & After' adjustments I made. I was trying to minimize the reflections and boost the details that were hidden in the shadows. Oddly, when I ran the initial alignment using the adjusted photos on the 'Low' setting, everything worked splendidly. I then went through the remaining steps of creating a dense cloud (success) then a mesh (success again) and finally created masks from the resulting low resolution model. My problems began after that when I used the masks and started over on the 'High' setting.

Photoscan just literally finished a few minutes ago creating the dense cloud model from the sparse cloud. I've cleaned up the model a bit as it had sections where the surface was lifted and hovering far away from the remainder of the model. I've included screenshots of the dense cloud model approximating the positions I stood at to photograph the original for comparison.

I guess I'm very linear in my logic, but I thought if it worked well on a low quality setting, that it would work just as well on a high setting, giving me far more detail in places like the face, the locks of hair, toes, fingers, etc.