v0.8.0

MAP-Tk v0.8.0 Release Notes

This is a major release of MAP-Tk that provides both new functionality and
fixes over the previous v0.7.2 release. However, MAP-Tk is still considered
pre-release software and subject to significant API changes.

The primary change in this release is a major refactoring of MAP-Tk into
two repositories. There is a new repository called Vital on which MAP-Tk
now depends. Vital now contains core framework constructs from MAP-Tk
while the MAP-Tk repository contains plugins, tools, the GUI application,
and some SfM specific helper functions. This split will allow easier
sharing of the core framework across KWIVER projects. Some core features
from the Vibrant repository (better logging and config parsing) were also
integrated into Vital, and MAP-Tk benefits from those.

As part of the refactoring MAP-Tk also moved to C++11. C++11 provides most
of what MAP-TK relied on Boost for. Thus, the Boost dependency has been
removed and replaced by standard library calls where available. For other
uses of Boost (e.g. filesystem, program options) we now rely on KWIVERSYS.
KWIVERSYS is code provided by Vital and original developed to support CMake.

The move to C++11 means dropping support for some older compilers. Visual
Studio 2013 and above is now required on Windows. Some versions of GCC with
only partial C++11 support are still supported. In particular, GCC 4.4.7 is
supported to allow builds on CentOS 5/6; however, GCC 4.4.7 still requires
Boost in some areas to address standard library deficiencies.

Using the new, more powerful config file parsing in Vital the default MAP-Tk
configuration files have been broken up into smaller files which reference
each other and use namespaces for readability and reuse. The default config
files are now installed and shared by the GUI and CLI tools.

There are numerous additional changes, especially in the GUI visualization
application. These changes are described below.

Updates since v0.7.2

MAP-Tk Library

• Moved a large number of core data types and abstract algorithms to a
  new dependent project name Vital. Vital will provide a common framework
  for logging, configuration, and core data types to be shared across
  KWIVER projects.
• Changed the camera and camera_intrinsic classes to act as higher level
  containers. Each now has an abstract base class and a "simple" concrete
  class. The templated derived classes (e.g. camera_) have been removed.
  The simple classes use the double type internally.
• Changed the camera to contain a shared pointer to a camera_intrinsics
  object rather than an instance. This allows cameras to share common
  intrinsics.
• Moved some common functions like project() and depth() into the camera
  base class to allow usage without down-casting.
• Changed the canonical transform function to set the orientation
  based on the eigenvector of the landmark point cloud. The cameras are
  now only used to set the sign of the up vector (Z-axis).
• Improved the Necker reversal transform so that the residual error
  is now kept low after the transform and the transform is now involutory.
  That is, applying the reversal a second time will return to the previous
  state.
• Removed required dependency on Boost. Replaced Boost with equivalent
  functions in the C++11 standard library when available. When not available
  kwiversys (found in Vital) provides equivalent constructs. The exception
  is that the unit tests will require boost::random for some older C++0x
  compilers without std::random.
• Applying transformations now better preserve additional data associated with
  landmarks (e.g. color, number of observations).
• Added a new utility module with functions to extract feature colors from
  an input image (and apply to relevant features in a feature track set), and
  to compute landmark colors from a feature track set with color information.
• Refactored some helper functions into epipolar_geometry.h for converting
  between fundamental matrices, essential matrices, and cameras, and for
  computing inliers to a fundamental matrix within feature matches.

Configuration

• The sample config files for the tools have now been refactored. Reusable
  default config files for each algorithm are now provided in the "config"
  directory and example top-level config files are provided in the "examples"
  directory, organized by data set. The example config files include the
  algorithm config files using the new "include" directive. Algorithm config
  files may include other algorithm include files.
• Config files are now installed with the application and can be found at
  runtime anywhere on the configuration search path. The configuration
  search path can be extended with the KWIVER_CONFIG_PATH environment
  variable.

Core Plugin

• Updated triangulate_landmarks to set the number of observations on each
  computed landmark.
• Improved initialize_cameras_landmarks to allow for incremental
  initialization that optionally runs an occasional bundle adjustment
  and can remove landmarks/tracks that still have larger error after
  bundle adjustment. It also applies Necker reversal and runs another
  bundle adjustment to test if the reversed solution has lower error.

Ceres Plugin

• Added an option to the bundle adjust algorithm to control how camera
  intrinsics are shared across cameras. The default auto sharing mode
  optimizes a separate set of intrinsics for each camera_intrinsics object
  shared amongst the cameras. Additional options allow forcing each camera
  to have its own independent intrinsics and forcing all cameras share
  one common set of intrinsics.
• Modified bundle_adjust to preserve additional data associated with landmarks
  (e.g. color, number of observations) during optimization.

VXL Plugin

• Updated triangulate_landmarks to set the number of observations on each
  computed landmark.
• Added an initial implementation of fundamental matrix estimation using VXL.

Visualization Application

• Added ability to execute some computations (for now, computing and applying
  basic transformations, applying bundle adjustment to refine the solution)
  from within the GUI.
• Added ability to export cameras and landmarks to (respectively) KRTD and PLY
  files.
• Added ability to visualize locations of corresponding feature points on
  adjacent frames to the camera view. This may be helpful as an additional
  means of visualizing camera motion when there exists a spatial correlation
  between adjacent cameras (e.g. frames extracted from continuous-shot video).
• Added "match matrix" view, allowing the number of corresponding feature
  points for arbitrary frame pairs to be visualized as a "heat map" with
  various Visualization options.
• Added ability to load imagery directly, and without requiring that cameras
  are loaded.
• Added ability to visualize landmarks by "true color", elevation, or number
  of observations, when relevant data is available.
• Added ability to selectively display ("filter") landmarks based on values of
  available data (e.g. elevation, number of observations).
• Added ground plane display to world view.
• Added ability to view camera frame image, projected to the ground plane, in
  the world view.
• Added ability to toggle world view between parallel and perspective
  projections, and to reset the rotation of the same to one of several basic
  presets (top, left, etc.).
• Added ability to change the background color of the views.
• The GUI user manual can now be accessed from within the GUI.
• Added an "About" dialog including information about the version number,
  acknowledgements, license, and build information.

OCV Plugin

• Images are now mapped so that the first channel of the vital::image view is
  red (RGB) rather than blue (BGR).

SketchUp Plugin

• Added a new plugin (Ruby scripts) for the SketchUp 3D Modeling software
  (http://www.sketchup.com/) that allow direct import of MAP-Tk outputs by
  loading the MAP-Tk config files or individually loading KRTD or PLY files.
  Once loaded, the cameras are presented as "Photo Match" pages which allows
  the user to build models aligned to images.

Tools

• The track features tool now extracts feature colors and writes them to the
  track file.
• The bundle adjust tool now computes landmark colors from feature colors and
  writes them to the landmark PLY.

Fixes since v0.7.2

OpenCV Plugin

• Fixed an issue with some parameters to draw_tracks and analyze_tracks
  in which the get_configuration function was always returning the default
  parameters, even after other values were set.

Visualization Application

• Several performance bottlenecks that limited the rate at which "slideshow"
  mode could switch frames have been identified and mitigated.
• Fixed bug where newly loaded landmarks would be visible in the world view
  even if landmarks were disabled.
• Fixed bug that caused feature tracks to be "misaligned" if a camera could
  not be loaded.