Microscopic Imaging and Masks for Insects

[linden]

Hi!

I'm trying to create a 3D model of a mosquito from photos under a microscope. Since I'm using the "turn-table" photogrammetry method, I took a back-lit photo for every front-lit photo to act as a mask.

Total I have 100 front-lit photos and 100 back-lit photos (masks).

I'm trying this workflow (https://agisoft.freshdesk.com/support/solutions/articles/31000158967-aligning-turntable-photos-with-background-suppression-from-single-mask-in-agisoft-metashape) where I import the masks and use them during the alignment step.

I read that applying the mask to the key point needs a mask on every photo, which is what I have, so I started with this option. After running the alignment, only 7 of my photos/cameras aligned.

So next I tried applying the mask to tie point where I read it works better with only a few representative masks so I choose 6 out of the 100 photos to have masks. After running this alignment, only 3 of my photos/cameras aligned.

To test and make sure I have enough overlap between the photos I tried running the alignment without masks and 100 of the 100 photos/cameras aligned, so the issue is using the masks.

I would greatly appreciate any feedback on whats's going wrong or if I should be using the masks at a different step of the process.


Thank you!
Linden

[Paulo]

Hi Linden,

are u sure the masks are correctly generated? With Show Masks in Photos Pane, you should see  the central part (insect) of each masked image in white and surrounding in black as in following:

Hope this clears up your doubts....

[linden]

Thank you Paul, it looks like I forgot to invert my masks which makes sense why my photos/cameras did not want to align. After inverting the masks all 100 aligned.

_____________

Although I have another issue:

I ran the photos through again (with the correct masks) and the output is not great. I know the detail on the antenna and mouthparts is tiny but I was really hoping the masks would help save it.

I've attached screenshots of my values for each step of the process, if anyone could offer insight into what I can change to get a better result, I'd really appreciate it. 

I also have a screenshot of what I get when using the same values but not the masks.


Thanks!
Linden 

[Paulo]

Hi,

maybe you can try building mesh directly from depth maps using mld filering....