[question] Trouble using know camera positions #950
Comments
How many features were detected for your images?
missing matches. Here is a summary of the Issue discussion https://github.com/alicevision/meshroom/wiki/Using-known-camera-positions |
|
The image set reconstructs very well with the default pipeline. We detect 2-5k features per image. I had not seen the wiki page or the discussion in #829 yet, I will have a look at those. Our main goal here is to get models with correct size, so perhaps the easier solution is to transform the pointcloud or mesh externally after Meshroom is done, to avoid messing too much with the Meshroom pipeline. Perhaps our positions are not as accurate as Meshroom needs them to be. |
|
You can use Meshlab for scaling https://www.youtube.com/watch?v=6psAppbOOXM The next release will allow scaling a model to real world size using CCTAGs. |
|
I will scale the model using the known camera poses and the poses from Meshroom using some simple Python code, but thanks thanks for the pointer. I am greatly looking forward to the next release, there are several interesting features I'm interested in. Is there a roadmap? I got my edited |
https://github.com/alicevision/meshroom/wiki#when-can-we-expect-the-next-binary-release |
…ion pull request #950 (#1180) * According to Meshroom issue #1179 (#1179), add the describer type "tag16h5" to the following modules: - ConvertSfmFormat (e.g., to be able to export the 3D AprilTag positions in a human-readable format as .sfm, or to see only the AprilTag marker positions in the 3D view via .abc) - FeatureExtraction (to be able to detect AprilTag markers from the tag16h5 family) - FeatureMatching (to be able to match AprilTag markers) - SfmTransform (to be able to use AprilTag markers, e.g., for the auto_from_markers transform) - StructureFromMotion (to be able to compute the 3D positions of AprilTag markers) * Added a new input to sfmTransform: markerDistances, which is a pair of marker IDs associated with the distance between them. Added a corresponding new transform: from_marker_distances, which scales the model according to the given distances between pairs of markers. Added another transform: auto_from_markers, which uses the existing markers parameter (ignoring their x,y,z positions) and applies the auto_from_... function only based on these given markers. The latter transform can, e.g., be used to align a set of markers with the ground plane. * Revert "Added a new input to sfmTransform: markerDistances, which is a pair of marker IDs associated with the distance between them." This reverts commit ed87c68. Co-authored-by: jarne <jarne@ieee.org> Co-authored-by: Fabien Castan <fabcastan@gmail.com>
Describe the problem
I have a set of images with accurately known camera poses. I'm trying to force these poses in Meshroom, but having some issues. I've followed the instructions from this post, but the StructureFromMotion node crashes when I try to run it using a modified ´cameras.sfm´.
In summary I do the following
cameras.sfmby setting the folders to empty"locked": "0"to"locked": "1"fordata["intrinsics"]and for alldata["poses"]"rotation"and"center"for all posesThe dataset computes fine with default settings.
Log
When running the SfM node with default settings
When running the SfM node with modified
cameras.sfmDesktop (please complete the following and other pertinent information):
Version 2019.2.0
Windows 10
Python 3.6.6
The text was updated successfully, but these errors were encountered: