Agisoft Metashape

Agisoft Metashape => General => Topic started by: tjh628 on February 18, 2020, 08:44:35 PM

Title: Turntable Modeling of large stone blocks - Invalid Points?
Post by: tjh628 on February 18, 2020, 08:44:35 PM
So I've been trying to do some modeling with Metashape 1.6.1 Standard of large, carved stone blocks of different shapes from an archaeological site. The blocks were set up individually on a turntable and shot with a stationary camera, then rotated slightly, etc. About 15-16 shots were taken from an oblique angle of the top, and then the stone was flipped and the process repeated, for a total of 30-33 photos for each block. I do a manual mask for each photo using the Intelligent Scissors tool as part of Metashape to remove the background. A few of these blocks process just fine, producing great models. Most, however, are not working out, and only some of the photos are aligning at all.

I didn't take the photos myself, unfortunately, and some of them are less than ideal: a few have some depth-of-field blur or possibly motion blur, they are in .JPG format rather than RAW or TIFF, and on some of the blocks, one or two sides are nicely smoothed and fairly uniform in color, so texture is an issue. But the photos that are not aligning don't seem to be related to texture. The "Estimate Image Quality" option for each chunk also returns values generally between 0.7 and 0.9 for most of the photos - I disable any photos below 0.55 or so. I've tried all of the combinations of different settings in the Align Photos step, but have had little success. I can sometimes get a few more to align if I run the whole photo set and then individually align photos that didn't get aligned as part of the larger step, but that's rare.

When I look up "View Matches" to see if I can troubleshoot why the photos might not be aligning, all the photos (generally) have a few dozen to a few hundred points matching up with the photos adjacent to them in the sequence, descending in quantity for the photos further away (often there are several photos with matching points in the single digits). For the photos that aren't aligning, however, all or almost all of these points are "invalid." The thing is, I can see the photos and the points, and they are matching up accurately! So I don't understand why they aren't considered "valid" matching points by the program.

While taking better photos might help in some cases, because of where the stones are from and the effort involved in getting to them, taking them over again will be a huge undertaking, and I obviously want to exhaust my options before I resign my team to that fate. So here's what I'm hoping someone can help with: (1) What makes Metashape decide some matching points are "valid" and others "invalid?" (2) Is there a way to loosen the tolerance of valid vs. invalid points, since I can see they do match up accurately, to increase the success rate of the modeling? And (3) Is the problem really with the photos, or is there something else I can try (and would there be any tools available in the Professional version that might help)?

Thanks - I'd really appreciate any help I can get.
Title: Re: Turntable Modeling of large stone blocks - Invalid Points?
Post by: Mak11 on February 18, 2020, 09:05:12 PM
The low number of shots taken indicates a lack of overlap IMO ( 30-33 is way too low especially if the subject doesn't have many distinguishable features).

Title: Re: Turntable Modeling of large stone blocks - Invalid Points?
Post by: tjh628 on February 19, 2020, 08:36:27 PM
Really? I would have thought that would be enough. Good to know that it's not!

Although it only partly answers my question about invalid points. The program is still correctly identifying matching points between photos but categorizing them as invalid. Is taking more, better photos the only way to fix this?