[bug] Invalid camera reconstruction for symmetrical buildings #605
Labels
bug
for actual bugs (unsure? use type:question)
do not close
issue that should stay open (avoid automatically close because stale)
feature request
feature request from the community
type:question
Describe the bug
We are trying to map a building which has many similarities, symmetrical if you want to keep it simple. One of the main problems we face is invalid camera reconstruction. As a clear example a screenshot. What happens is the left side tower of the church is reconstructed on the right side.
To Reproduce
We can make a dataset of the project available.
Expected behavior
I understand it is sometimes impossible to differentiate between ambiguous images. This is fine, even in a human situation it is impossible to know without context. Still in this specific case images are available that clearly are zoomed and have left and right overlap. What I would like to say about a reconstruction: "Meshroom, this camera reconstruction is wrong." or ultimately: "Meshroom move and pan the camera, and attempt a local beam reconstruction for this single image."
Suggestions to improve the situation with settings are obviously also appreciated. Our minimum input track length was already increased. With Minimum Input Track Length to 6, and Min Observation For Triangulation to 4 the results look optically fine. I could now comment on the fact that this is still problematic for bigger projects. Absolute minims are; Minimum Input Track Length to 2 and Min Observation For Triangulation to 3 (a lower number results in mirroring of the building).
Desktop (please complete the following and other pertinent information):
The text was updated successfully, but these errors were encountered: