Lately I've been working on a full body scanner with 63 cameras, and I keep running into this problem.
I've gotten some really good models with it but some days I just can't get a full model for the life of me. The biggest problem I'm having is with alignment, below I have two examples.
I'll start with Ashley. Previously his model was being broken up into about 9 components, but after tweaking some settings I've gotten it down to 3:
Ashley is wearing black pants and a black hat which tends to be a bit tricky, but this scanner needs to be able to handle blacks so this is a good test.
Next up is John, who's also wearing black pants but has a bright orange shirt on. I took the picture using the same exact cameras, rig, and lighting, and ran it through RC with the same settings, as well as cleared application cache (I cleared the cache before running Ashley through as well). This time however it gets split into 9 components.
So my question is, what's the difference between the two shoots to justify such a radical change in results? I'll post all the relevant files which, while I was uploading them, gave me the perfect example of how inconsistent this problem is.
When I took the screenshot of Ashley's alignment results I had 3 components with 34, 22, and 6 cameras, for John I had 9 components. I then reran Ashley so I could get a save file to upload, and in doing so I got 2 components with 40 and 23 cameras (Ashley2Com.zip). I then ran Johns again and got 11 components (JP.zip). I restarted reality capture and ran Ashley again to get 3 components with 23, 34, and 6 (Ashley.zip). I restarted again and ran John to get 10 components (JP10Com.zip).
I also posted the alignment settings, Alignment.xml, and the image files. Ashleys images are in Ashley21418.zip, and Johns are in JP21518.zip.
Any help you can provide would be much appreciated!
Files on Google Drive, let me know if you'd prefer me to upload them somewhere else:
Please sign in to leave a comment.