Hi Devs,
Thanks for all your hard work on this project.
I have been attempting to process a large dataset (between 4 and 64k images) on an Azure VM with 2TB RAM and 4TB SWAP.
The most recent test failed due to an overflow on the Ceres Solver num_nonzeros_ variable. It uses a 32-bit integer for storing some values. Attempts to support a 64-bit int have been deemed too challenging to pursue. 32-bit integers used in block_sparse_matrix.cc · Issue #412 · ceres-solver/ceres-solver · GitHub
ODM offers a split-merge option for large datasets which produces multiple meshes as an output.
This request it to merge these textured meshes and offer them to the end user as an output just like in non split-merge processes.
Many thanks,
Declan
3 Likes
We are trying to find someone able to help us get that work on Ceres done 
Have you tried using MeshLab to merge the split models for the time being?
2 Likes
Cool. Thanks for opening this request. We should start by just returning the available meshes, much as we do for the aggregate products. I have opened an issue here to reflect this:
Figuring out aggregation / merging of the returned meshes will need to be a later step, but at least returning the individual meshes would give folks something to work with.
3 Likes
We did this with models from adjacent non split-merge meshes. It is a pain when we are looking for hi res as meshes need to be joined with no cracks.
But I will run a test using a split-merge and see what the available meshes look like.
1 Like
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.