jmbuena / img_align_lib.python Goto Github PK

We can check if the image is too small at a given pyramid level and stop at that level

Save the result of tracking in an xml file. It will be useful later to evaluate performance and compare different algorithms

MotionModel, ObjectModel and OptimizationProblem are abstract base classes (a.k.a. Java interfaces). I don't know if I have implemented right the "C++ protected attributes" in python.

Implement in the test directory the classes needed for automated test execution. All the configuration should be done with .xml files and should allow to add new ObjectModel, MotionModel, CostFunction or Optimizer sub-classes in the atutomated test execution framework.

This algorithm is an Inverse Compositional one that learns the pseudoinverse of the Jacobian instead of computing it as the pseudo inverse of the analytic Jacobian.

The implementation should:

• Have a new derived class from OptimizationProblem called, for example, JurieInvCompProblem.py
  which takes the training image(s) in the constructor and learn the Jacobian pseudoinverse (and the Jacobian?)
• Make a new test script code/jurie.py (here the training image(s) should be passed to the new optimization problem).

Ideally should be a Decorator over any ObjecModel that has an image in the template.
If not possible to get a Decorator it can be just a subclass (or superclass) for ModelImageGray and others.

See paper:

Robust Tracking in Low Light and Sudden Illumination Changes
International Conference on 3D Vision (3DV), 2016
Hatem Alismail, Brett Browning, and Simon Lucey

Data as:

• Time per iteration
• Number of iterations
• Time per frame
• Other?

Is interesting for the evaluation of algorithms.

Ideally should be a Decorator over any ObjecModel that has an image in the template.
If not possible to get a Decorator it can be just a subclass (or superclass) for ModelImageGray and others.

See paper:

Robust 3D Tracking with Descriptor Fields
Alberto Crivellaro, Vincent Lepetit
CVPR 2014

I'm not sure that this conditions are fully right