flow_library

Pure python library providing tools for optical flow computation.

Features:

• Read and write optical flow files and disparity files
• visualize optical flow fields
• interactive GUI to visualize optical flow files
• compute error measures
• handle optical flow datasets: MPI Sintel, KITTI 12, KITTI 15, Middlebury

Installation

Clone the repository and install the necessary requirements. In order to use this library, add it to the PYTHONPATH environment variable.

git clone [email protected]:cv-stuttgart/flow_library.git
cd flow_library
pip install -r requirements.txt
export PYTHONPATH=$PWD

Finally, the library is able to manage dataset filepaths and automatically detect groundtruth flow files if the DATASET environment variable is set to the folder containing the desired datasets. The datasets folder should be structured as follows:

$DATASETS
    > kitti15
        > testing
            > image_2
        > training
            > flow_noc
            > flow_occ
            > image_2
    > mpi_sintel
        > test
            > clean
            > final
        > training
            > clean
            > final
            > flow
    > ...

Optical Flow GUI

To use the GUI, right-click an .flo or .png flow file and choose Open With > Other Application. There, open the dialog to search for a custom executable, navigate to the flow_library folder and select flow_show.py (for Windows users: flow_show.bat)

In the GUI, you can interactively select the scaling factor (default is the maximum flow vector length) and select the visualization type. The library tries to automatically detect a groundtruth flow file and shows the average endpoint error (AEE) and the percentage of bad pixels (Fl error). Use the arrow keys to traverse the current directory.