Faldoi minimization strategy

This respository contains the source code of Faldoi algorithm described in Faldoi. In addition to all the functionals explained in the article this code has another functional implemented, the one from Occlusion estimation.

Compiling

To compile type

        mkdir build
        cd build
        cmake ../src -DCMAKE_BUILD_TYPE=RELEASE
        make -j4

in the directory where this file is located. The compilation of the source code provides three executables:

1. sparse_flow - Convert the .txt files from match_cli to a sparse flow

2. local_faldoi - Estimates an optical flow from a set of initial sparse matches.

3. global_faldoi - Estimates a global minimization using as initialization the output from local_faldoi.

Executable - Usage

Usage: ./sparse_flow list_matches.txt colums rows out.flo

Usage: ./local_faldoi i0 i1 in.flo out.flo sim_map.tiff [options...]

options:

•        -m (0)  Changes the functional (check aux_energy_model.h)
  

•        -wr (5) Radius value 5 - patch 11x11 
  

Usage: ./global_faldoi I0 I1 input.flo out.flow

options:

•        -m (0)  Changes the functional (check aux_energy_model.h)
  

•        -w (5)  Number of warpings   
  

Script - Usage

There exist some python scripts for executing all the code at once. To use them, go to scripts_python folder. In the folder example_data, there is an example of images to use.

fast_faldoi

To use the python script (fast_faldoi.py) that makes all the process, first you need to do the following:

• Compile the executables through the commands above
• Put the excutable deep matching from DeepMatching: Deep Convolutional Matching into build directory from DeepMatchings

Usage: python fast_faldoi.py i0 i1

fast_sift

To use the python script (fast_sift.py) that makes all the process, first you need to do the following:

• Compile the executables throughthe commands above
• Put the excutables sift_cli and match_cli from Anatomy of SIFT from IPOL into build directory from SIFT

Usage: python fast_sift.py i0 i1

fast_faldoi_occ

To use the python script (fast_faldoi.py) that makes all the process, first you need to do the following:

• Compile the executables through the commands above
• Put the excutable deep matching from DeepMatching: Deep Convolutional Matching into build directory from DeepMatchings
• Create a txt file with the path of the images in the following order: I_0, I_1, I_{-1}, I_2

Usage: python fast_faldoi_occ.py list_images.txt