You are in right place, if you want to use graph matching procedure on already computed similarity / cost matrix.
+ No need to implement own features
- Limited sequences size. Up to the size of the matrices that fit in RAM.
Change the respective parameters in the config_cost_matrix.yaml to account for your cost matrix.
Then, run:
./run_localizer_cost_matrix_based pathto/config_cost_matrix.yaml
To create the cost matrix, please visit create_cost_matrix
The code knows how to read the matrix from a .txt file, stored as matrix.
First line in the .txt should specify number of rows and cols that the matrix has.
For more details, please check database/cost_matrix_database.h.
Warning this code expects the costs to be cosine distance costs, e.g. the higher values correspond to the better match.
If your costs have different meaning, e.g. the lower the cost the better the match, then you need to either invert your costs before passing in to a program or derive from the class CostMatrixDatabase a class with your own implementation of getCost function.
Upon finishing the code should provide the file matched_path.txt with image associations.
Additionally, it should generate an image with costs and overlayed path (red/blue) as well as expanded nodes (green).