Agisoft Metashape
Agisoft Metashape => General => Topic started by: holgar on February 21, 2020, 01:45:06 PM
-
Hello,
I am using Agisoft Metashape/Photoscan since many years. But I have allways had troubles to align my Images I make on my turntable. I take a lot of care to produce good images:
- Checking for diffuse Light
- Checking for plain background
- Checking for good sharpness and resolution
- Checking good overlap (100-200 Images per Object)
- I even use Crosspolarization techniques to prevent reflections.
- On small objects I sometimes use focus stacking (works sometimes. Sometimes not. Capture Reality works allways)
No matter what I do the Alignment fails very often on many images. Many times it is half the Cameras of one side (when the object rotated 180 degree.).
This is my main Problem with Metashape. I have tried Capture reality a few times with my failing Objects. And Capture reallity allways does a perfect Job aligning.
Also Capture Reality produces way more Tie Points than Metashape ( I set Tiepointlimit in Metashape allways to Zero).
So I wonder if there is something that can be improved in Metashape to make aligment more stable? I dont want to use masks (which helps sometimes). This adds an extra step that I want to avoid.
I would really like to keep using Metashape as I like it a lot. Also the pricing of Capture Reality prevents me from using it for my private Projects. Also I like some features and the Interface of Metashape more.
Please let us know if there is anything that can/will be improved in this regard.
Thanks a lot for the great work and the frequent updates. Greetings, Holger
-
Hi Holgar,
when you are using a robotic turntable it might be possible to use a pre oriented project, which has a good alignment in the past, to import your photos in and they are all pre oriented this way. Perhaps worth a try.
Regards,
Kiesel
-
That’s actually an idea I will try. Thanks. But still there must be something that could be improved on the software side. I can not scan everything with the same amount of pictures. Somehow metashape does not find as many keypoints and tiepoints as it should. Also it finds to many keypoints in areas where it shouldn’t. So I hope there will be improvements on that.
Capture reality find way more points in relevant areas. Sometimes a multitude more than metashape. And therefore also the dense cloud results in capture reality is also much better.
-
Hi Holgar,
you can also put a sheet with printed coded markers on your turntable (you can download one here: https://3dscanexpert.com/review-foldio360-3d-scanning-photogrammetry/ (https://3dscanexpert.com/review-foldio360-3d-scanning-photogrammetry/)) and also scale your model that way or use something with a good texture (newspaper, cork,...) to help Metashape in alignment step.
Regards,
Kiesel
-
Hi Kiesel,
That is not an option for me unfortunatly. I have to scan my Objects in the Air (with a Rig) to be able to scan all Sides. I usually Have a thin stick where my Objects are glued to or I use transparent Nylon cord to lift the object in the Air.
Here is an Example image of what I usually Capture ( I reduced the size):
-
I would highly suggest using this technique:
https://www.agisoft.com/index.php?id=49
Never had a single alignment issue in nearly 2 years.
Mak
-
I would highly suggest using this technique:
https://www.agisoft.com/index.php?id=49
Never had a single alignment issue in nearly 2 years.
Mak
Hi Mak,
thanks for suggesting this. I actually do this workflow and it works fine. But I think it should and could without that workflow which adds extra steps to the process. Capture reallity works without any of these steps. So maybe it could be improved in Metashape. When I have to do 20 Scans in one week it is a quite a hassle to make this extra step.
Yesterday I tried to capture my Photos with a Black background and than adjust the levels that the back is almost complelty black. Than I do Mask from Background with a Black image. That works good. The downside is that the Black background does effect the lighning and it is harder to have a uniform diffues Light without shadows.
I am producing highend Models for CG. So it is crucial to have good diffuse light..
-
Here is an Example I aligned in Metashape and Capture Reality. I used no masking in both applications. Its a very small object with a very narrow depth of field. So many blurry areas. But I took egnough photos that all areas are covered well.
Metashapes only aligns 40 Cameras and produces 13k Points. Capture Reality algins all Photos and produces 150k Points. I wonder what are the differences in the Algorithm that makes Metashape fail here. I hope this can be improved.
-
One Idea that came to my mind. Maybe the problem is Metashape only considers Luminance Vallues (greyscal) for the Alignment process? I imagine it could be more precice using full Color Vallues? But i am not an expert in this...just a thought.
-
Also a Question: Is there any way I can adjust anything regarding the Feature detection? Any hidden parameter that I can change?
-
I read about the TWEAKs section in the Preferences. Is there parameters I can use to adjust the Feature detection maybe? Is there a full documentation of these Tweak parameters? I cant find anything in the documentation.