search

openphotogrammetry / meshroomcl

MeshroomCL: An OpenCL implementation of photogrammetry with the Meshroom interface
Other
169 stars 8 forks source link

Augment Reconstruction? #14

Open WBoode opened 3 years ago

WBoode commented 3 years ago

So I have used the Augment Reconstruction quite a bit on the default Meshroom, however on the OpenCL pipeline, this will currently not work. Is the inclusion of Augmented / Live reconstruction being considered at the moment?

revisionarian commented 2 years ago

Hi @WBoode, you're correct that the StructureFromMotionCL node will not currently work with the "Augment Reconstruction" feature. However, in the latest release of MeshroomCL (version 0.8.0) you can select a photogrammetry pipeline that uses the default Meshroom (non-OpenCL) StructureFromMotion node, and then connects to the OpenCL MultiviewStereoCL node. You can select this pipeline as the menu item "Photogrammetry (OpenCL MVS)" under the File → New Pipeline menu.

With this pipeline, you can use the "Augment Reconstruction" feature in MeshroomCL the same as you did with Meshroom. Let us know if you have any more problems!

Sleepiejak commented 2 years ago

Just recently got into Photogrammetry and for me the whole reason to use the OpenCL version was because i run an AMD graphics card. and im having an issue when it comes to the feature matching is looking for a NVIDA card for CUDA. is there anyway around this? for i have to add the OpenCL version of the node instead. I just thought of that the auto might still be set to the CUDA one by default.

revisionarian commented 2 years ago

Hi @Sleepiejak, I am confused about what problem you are encountering with MeshroomCL. When you run MeshroomCL, the default photogrammetry pipeline will have 5 nodes (CameraInit, StructureFromMotionCL, MultiviewStereoCL, MeshFiltering, TexturingCL). None of these nodes use CUDA in any way.

If you use the File → New Pipeline menu to change from the default pipeline to the "Photogrammetry (OpenCL MVS)" pipeline, then there will be 8 nodes, one of which is FeatureMatching. Is this the pipeline that you are using?

The FeatureMatching node does attempt to detect the presence of CUDA, but it doesn't require that CUDA exist. If there is no CUDA, then the FeatureMatching node will use a CPU-based algorithm. Could you please post the error message (or the node log file) that is occurring when the FeatureMatching node fails due to looking for CUDA?