block-lma-awf-expts

View source
block-lma-awf-expts
  • Contributors
    1. Loading...
Author Commit Message Date Builds
1 commit behind default.
Andrew Fitzgibbon
Add comment to help diagnose assert failure
Andrew Fitzgibbon
Rank was incorrectly computed before backsolve
Andrew Fitzgibbon
Handle rank correctly
Andrew Fitzgibbon
Triplet needs Index, not StorageIndex. Derive TripletArray from its container.
Andrew Fitzgibbon
Remove general InputType example
Andrew Fitzgibbon
int->Index, other cleanups
Andrew Fitzgibbon
int -> Index
Andrew Fitzgibbon
Use fewer points in debug builds
Andrew Fitzgibbon
remove tictoc
Andrew Fitzgibbon
Merge
Andrew Fitzgibbon
WIP trying to compile merge with suitesparse blas etc
Andrew Fitzgibbon
Merge, not yet compiling
Andrew Fitzgibbon
WIP: awf
Andrew Fitzgibbon
Obey NDEBUG, improve typedef names.
Andrew Fitzgibbon
Add EIGEN_DEBUG_ASSERTS
Andrew Fitzgibbon
Fix Index types.
Andrew Fitzgibbon
Fix PermutationType
Andrew Fitzgibbon
Obey EIGEN_NO_DEBUG
Andrew Fitzgibbon
add size() to TripletArray
Andrew Fitzgibbon
add fixme
Andrew Fitzgibbon
Speed up matrixQ() for fixed-size matrices. This is ~100x speedup for a tested 5x3 matrix.
Andrew Fitzgibbon
Block<Diag<2,1, Fixed>,Dense>
Andrew Fitzgibbon
Points 2K, Block<Diag<2,1>,Sparse>
Andrew Fitzgibbon
Points 2K, Block<Sparse,Sparse>
Andrew Fitzgibbon
Add timers
Andrew Fitzgibbon
Bump N to 500K
Andrew Fitzgibbon
Remove unneeded temporaries
Andrew Fitzgibbon
clean up temporaries.
Andrew Fitzgibbon
Merge eigen/eigen. Fixed BockSparseQR to new definition of assign_op
Andrew Fitzgibbon
WIP: Memory errors in PermutationMatrix assignment. Updating to 3.3 to test.
Andrew Fitzgibbon
Generate some output to follow test progress
Andrew Fitzgibbon
Better sparse matrix size estimate
Andrew Fitzgibbon
Fix Q definition for rectangular matrices. Make work with fixed-size matrices in the blocks..
Andrew Fitzgibbon
Implement Product for BlockSparseQR matrixQ. This also required rationalization of Index vs StorageIndex. Permutaton matrices should really operate on the same type as the arguments of coefRef(i,j).
Andrew Fitzgibbon
Convert macro to function to allow breakpoint to be set within
Andrew Fitzgibbon
WIP moving to BlockSparseQRMatrixReturnType
Andrew Fitzgibbon
Improve readability
Andrew Fitzgibbon
Add makeCompressed
Andrew Fitzgibbon
Fix landscape case
Andrew Fitzgibbon
Add another test
Andrew Fitzgibbon
Spill some xprs to temporaries to avoid assert fail in insertBackByOuterInner. There's a deeper bug somewhere but this fixes it for me right now
Andrew Fitzgibbon
Fix for rectangular matrices
Andrew Fitzgibbon
Add leftCols
Andrew Fitzgibbon
Add comment
Andrew Fitzgibbon
Need to reset Jacobian in case sparsity pattern changes per iteration.
Andrew Fitzgibbon
Another speedup, several tightenings of types (e.g. PermutationMatrix must be over index type (the return type of row()), not StorageType)
Andrew Fitzgibbon
Small readability fixes
Andrew Fitzgibbon
Better comments
Andrew Fitzgibbon
Add to small tests group
Andrew Fitzgibbon
Add SparseMatrix::Iterator, to iterate over all nonzeros. The two-line idiom previously recommended is error-prone and inconvenient.
Andrew Fitzgibbon
Significant memory-use improvements to give Economy-size decomp.
Andrew Fitzgibbon
Add add_if_nonzero for convenience
Andrew Fitzgibbon
Bugfix
Andrew Fitzgibbon
Clean up headers
Andrew Fitzgibbon
Add TripletArray. Given that this is pretty much the recommended way to assemble triplets, it seems useful to encapsulate it.
Andrew Fitzgibbon
Add sparse matrix support
Andrew Fitzgibbon
Cleanups and comments
Andrew Fitzgibbon
Remove legacy LevenbergMarquardt implementation. There were two very similar implementations of LevenbergMarquardt, one in NonlinearOptimization, one standalone. This is potentially very confusing for those new to the library, so I thought it important to reduce this to a single implementation. Some forensics on the checkin history indicated that the standalone was newer. I ported some remaining differences from the old implementation (all but …
Andrew Fitzgibbon
Porting comments over from NonLinearOptimization
Andrew Fitzgibbon
Weaken test severity for iteration-count tests
Andrew Fitzgibbon
Backport sqrt_epsilon fix from legacy LM
Andrew Fitzgibbon
Allow optimization over arbitrary parameter vector types.
Andrew Fitzgibbon
Change ellipse_fitting to use BlockSparseQR with BlockDiagonalSparseQR. This is the QR analogue of the "Schur complement" trick in bundle adjustment
Andrew Fitzgibbon
Include guards (important), fix comments
Andrew Fitzgibbon
Merge
Andrew Fitzgibbon
Merge
Andrew Fitzgibbon
Small readbility fixes
Andrew Fitzgibbon
Make LM verbose
Sergio Garrido
add permutation handling, templatization and improve test
Sergio Garrido
Add internal solver templatization, column permutation handling, better checks and test
Andrew Fitzgibbon
Improve comment on cmakefile, which enables only one test
Andrew Fitzgibbon
Merge
Andrew Fitzgibbon
Andrew Fitzgibbon
Clean up sparse matrix construction, fix narrowing warnings.
Andrew Fitzgibbon
Infer type from container to avoid VS warnings.
Andrew Fitzgibbon
No unistd on windows. Not sure we need to pull down the base typedefs.
Andrew Fitzgibbon
Fix simple warnings
Andrew Fitzgibbon
Remove restriction that mat isCompressed(). This adds a copy if mat is column major, but is not significantly expensive, particularly as analyze is assumed to be called once when factorize is called many times, and factorize already involves a copy.
Andrew Fitzgibbon
Make StorageIndex available as a typedef for clients constructing Triplets.
Andrew Fitzgibbon
Put tests in subfolders to declutter IDE Visual Studio (and perhaps other IDEs) by default presents the Eigen workspace as a flat list of tests, including subtests. This patch places these in (IDE) subfolders with paths such as Official/test/sparseqr/sparseqr_1 There is no change to the on-disk layout, just the IDE.
Andrew Fitzgibbon
tmp
Sergio Garrido
add template for internal solver in blockdiagonalqr
Sergio Garrido
cleaning comments
Sergio Garrido
cleaning comments
Sergio Garrido
add ellipse_fitting test and required changes to lma
Sergio Garrido
Merged in block-sparse-qr (pull request #2)
Sergio Garrido
Merged in sparse-diagonal-qr (pull request #1)
Sergio Garrido
Created new branch block-lma
Sergio Garrido
Add BlockSparseQR
Sergio Garrido
Created new branch block-sparse-qr
Sergio Garrido
Add BlockDiagonalSparseQR
Sergio Garrido
Created new branch sparse-diagonal-qr