My students and I have begun a project using Slicer Photogrammetry to create 3D models of songbird skulls from natural history collection in the area. The platform has worked well for us with the largest specimens we’ve received on loan, which are about 55-60mm in maximum length. However, we’ve run into issues with our smallest specimens, which are ~25mm in maximum length. We’ve tried factorial combinations of troubleshooting fixes like taking up to 64 pictures per set of photos, increasing to 6 unique photo sets (instead of 5), and using a macro lens to make the specimen larger in the actual images while still being totally in focus. At best, the ODM model made a fragmented model with “bone fragments” floating in space, like so:
However, basically all of our ODM runs with this specimen have failed to assemble and gave an error saying that we need more photos with more overlap.
My question for the community is: What is the smallest size of specimen/object you’ve successfully used Slicer Photogrammetry to digitize? We will keep plugging away on larger specimens as best we can, but any idea of whether we need more photos, more photo sets, or if there’s a limit of how small this tool can go for handling our smallest specimens would be graciously appreciated. Many many thanks!
Photogrammetry doesn’t know anything about the scale. For small specimens the photography is the real battle, and two things matter most: the lens and the framing.
On the lens — skip the macro. Macros introduce geometric distortion (fisheye-type warping) that’s unacceptable for morphometrics. Use a good telephoto and zoom in on the specimen instead. You also want a lens you can stop down to a high f-number to keep the entire specimen within the depth of field; the tele lenses that do this well aren’t cheap (think the ones birders and wildlife shooters use).
On the framing — the specimen needs to be entirely in focus and fill ~80% of the frame. A small skull in the center of a mostly-empty frame is the classic failure mode: the feature matcher can’t find enough to lock onto and can’t separate the specimen from the background. That’s exactly what your “need more photos with more overlap” error is — ODM saying consecutive views don’t share enough matched features. Shoot in small angular increments with heavy overlap between adjacent frames.
One more thing working against you at this size: bird skulls are smooth and low-texture, so use diffuse, even lighting (kill any specular hotspots) against a plain background. With small, shiny bone, lighting is often what makes or breaks a run. In our 2023 paper, one extremely bleached mountain beaver skull was the one that consistently failed the reconstruction using the exact protocol for photography and photogrammetry.
Also photogrammetry within Slicer ecosystem is fairly niche. You might get better support and advise directly from the OpenDroneMap community forum. Browse those as well: https://community.opendronemap.org/
Thank you so, so much for such a speedy and detailed answer to our question!!! These suggestions have definitely given us some optimism and more things to troubleshoot to get these tiny skulls to work (hopefully). The issue I’m currently encountering is the tradeoff between the focal length of our lens and the depth of field, unsurprisingly. Our lens can handle 140mm of zoom and an aperture of 36 but, even when we have the camera as close as possible at the minimal focus distance (45cm), our depth of field is only about the same size as the specimen. Even then, the specimen is only occupying about 50% of the image. The depth of field should ideally be quite a bit wider, right?
We’ll keep at it and will cross our fingers. Thanks again for all your help and for supporting this phenomenal resource for us to use!
My experience with small specimens is limited. But two things you can try:
If you can’t get your specimen to occupy more than 50% of the frame, you can crop the black area so that more of the frame is the specimen.
To deal with limited depth of field, you can try focus stacking (take 3-4 pictures at different depths and stitch them to a single image). Although this is extremely tedious (and the main reason I never explored close range photogrammetry for small specimens).
