Forum

Author Topic: RMS Reprojection Error vs. Max reprojection Error  (Read 4620 times)

holgar

  • Jr. Member
  • **
  • Posts: 62
    • View Profile
RMS Reprojection Error vs. Max reprojection Error
« on: June 20, 2016, 11:00:03 AM »
Hello, I made scan with 577 Cameras. I did as always some filtering and applied a optimizes Cameras.
Now I have a much less RMS reprojection Error. But the Max reprojection Error got worse.
I wonder what is the reason here. Should I prefer the optimized version with less RMS reprojection or should I keep the unoptimized version with less Max Reprojection Error?


stihl

  • Sr. Member
  • ****
  • Posts: 410
    • View Profile
Re: RMS Reprojection Error vs. Max reprojection Error
« Reply #1 on: June 20, 2016, 01:16:08 PM »
I'd definitly go for the project with the lower RMS error.

The max error refers to outlier tie-points that are most likely irrelevant.

Got a screenshot of your tie-points?

holgar

  • Jr. Member
  • **
  • Posts: 62
    • View Profile
Re: RMS Reprojection Error vs. Max reprojection Error
« Reply #2 on: June 20, 2016, 01:18:42 PM »
I'd definitly go for the project with the lower RMS error.

The max error refers to outlier tie-points that are most likely irrelevant.

Got a screenshot of your tie-points?

Ok Thanks. I will stick to the optimized version than. Here is a screencap of the Tie Pointcloud. Its pretty dense still after filtering with 900k points. I guess that should be good ;-)
« Last Edit: June 20, 2016, 01:27:52 PM by holgar »

stihl

  • Sr. Member
  • ****
  • Posts: 410
    • View Profile
Re: RMS Reprojection Error vs. Max reprojection Error
« Reply #3 on: June 20, 2016, 07:50:49 PM »
That looks pretty good. Very good overlap because I see hardly any missing regions.

You could try and manually delete some of the floating outlier tie-points to clean it up further, but it's not explicitly necessary .

holgar

  • Jr. Member
  • **
  • Posts: 62
    • View Profile
Re: RMS Reprojection Error vs. Max reprojection Error
« Reply #4 on: June 20, 2016, 08:09:36 PM »
That looks pretty good. Very good overlap because I see hardly any missing regions.

You could try and manually delete some of the floating outlier tie-points to clean it up further, but it's not explicitly necessary .
Yes, I have already deleted some floating points. The only problem area is the hands as they are reflective.