OpenMVG Versions Save

OpenMVG "Paracheirodon innesi diamant" V0.8.1 is out.

It's recommended to checkout the master tree from the Git rather than download this archive.

What's new:

• Speed enhancement:
  • Incremental SfM pipeline: Looking for next image to resect is now done in parallel.
  • Matching of image collection: a bit faster thanks to a better usage of parallelism.
• Logging:
  • If graphviz is installed, view graphs are exported for the Image matching and the Global SfM processes.
• Connection with 3rd party application:
  • Extended the BAF ASCII output format for easy usage of OpenMVG SfM scene with 3rd party applications
  • Add a software to export undistorted images
• API:
  • each openMVG module now use it's own namespace "openMVG::module"
  • Remove the deprecated openMVG/bundle_adjustment module
  • Rewrite some samples to fit the new openMVG/sfm and openMVG/cameras modules
• Build (experimental):
  • (optional) openMVG can now use external Eigen & Ceres libraries.
  • OpenMVG can now be used as a third party library with Find_package(OpenMVG). (Prefer to use a local openMVG installation path: see BUILD)

Here the complete list of closed issues: https://github.com/openMVG/openMVG/issues?q=milestone%3A%22Putative+0.8.1%22+is%3Aclosed

• Move doc from LaTeX to a Sphinx one (ReadTheDocs)
• Calibration:
  • automatic computation of approximative focal length,
  • automatic detection of image groups that share common intrinsics,
  • estimation of the Brown radial distortion model during SfM,
  • add export to CMPMVS.
• better cross platform compilation.
• updated camera and bundle_adjustment wrapper.

Structure from Motion:

• Improved accuracy of the sequential chain (ACC12),
• Add a global SfM chain (ICCV13),
• Improved reconstruction export (visibility information)

Multiple View Stereovision:

• Enable large scale dense point cloud computation with project export to CMVS/PMVS.

Various:

• 7-dof rigid registration,
• Add a Linear Programming module and LInfinity norm optimization solver for geometric computer vision problems.

Paper implementation: “Multiple-View Geometry under the L infty Norm.” Authors: Fredrik KAHL, Richard HARTLEY. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 2008.

"Virtual Line Descriptor and Semi-Local Graph Matching Method for Reliable Feature Correspondence". Authors: Zhe LIU and Renaud MARLET, BMVC 2012.

"Global Multiple-View Color Consistency." Authors: Pierre MOULON, Bruno DUISIT and Pascal MONASSE. CVMP 2013.

"Global Fusion of Relative Motions for Robust, Accurate and Scalable Structure from Motion." Authors: Pierre MOULON, Pascal MONASSE and Renaud MARLET. ICCV 2013.

OpenMVG Core Modules

• Multiview:
  • Add an implementation of L1 rotation averaging: ICCV2013 "Efficient and Robust Large-Scale Rotation Averaging" Avishek Chatterjee, Venu Madhav Govindu (Indian Institute of Science).
• Matching:
  • Add support of HAMMING metric for binary descriptors (can be used in BruteForceMatching)
• SFM (general)
  • Colored calibration point cloud: reconstruction color is now an option
  • Reconstructions can be exported to support CMVS camera clusters computation
  • Make all related SfM serializations to disk faster
  • Improved SfMViewer (added camera cones display and keyboard navigation)
• GlobalSfM
  • More stability and improved multi-thread usage
  • Rotation averaging can be done thanks to the L1 or L2 back-end

Continuous integration Move from drone.io to travis.ci (Now GCC and CLANG compilers are tested)

Build/CMake UnitTest and Samples build can be set to ON or OFF, resulting in faster build if used in external project.

OpenMVG Core Modules

• Multiview: rotation averaging:
  • make L2 and L1 rotation averaging using the same input interface,
  • faster L2 rotation averaging. minor bug fixes.
• SFM (general) enable MultiThread BA, some code-refactoring.
• system introduce a timer class.

Continuous integration Add code coverage to continuous integration: 90% of openMVG core is unit tested -> https://coveralls.io/r/openMVG/openMVG

Build/CMake

• Use a new EXIF parsing library,
• fix for latest Mac OS clang.

OpenMVG "Paracheirodon simulans" V0.8 is out.

It's recommended to checkout master tree from the Git rather than download this archive

What's new:

• Introduce a SfM data container and a sfm core module.

The sfm module is highly flexible and can uses:

• abstract views (image metadata storage),
• abstract features & descriptors (image descriptions),
• abstract camera intrinsic interface (Pinhole, PinholeRadial1/3).

Sequential & Global SfM pipelines:

• have been rewritten to fit new data structures
• are now unit tested!
• export a HTML SfM statistics report.

The Global SfM pipeline supports now multiple focal groups (Stéphane Flotron).

• Features/descriptions:
  • OpenMVG AKAZE implementation (Romuald Perrot).
• Geometry:
  • Robust triangulation from known camera poses,
  • Camera frustum intersections.
• Ready to use demo & tutorial python scripts for launching OpenMVG SfM pipelines.

Improved data serialization and code base. Released camera trajectory accuracy evaluation code (regarding dataset with ground truth).