MatchingBenchmarking

A program that determines which matching algorithm or neural network best fits the needs for the Clew app.

Getting Started

Dependencies

• opencv-python
• numpy
• scipy
• pandas
• matplotlib
• torch
• pickle
• protobuf
• progressbar

File Structure

Main files

• main.py: The main file that runs the program.
• FirebaseWrapper.py: Contains the FirebaseWrapper class that handles downloading data from Firebase Storage and extracting session data from it.
  It runs these scripts before collecting the session data:
  • sync_firebase_storage.sh: A bash script that syncs the Firebase Storage bucket with a local directory.
  • untar.sh: A bash script that untars all the tar files into correctly nested directories.
• SessionGenerator.py: Takes the session data generated by FirebaseWrapper.py and creates a pickle file containing all the sessions.
• Benchmarker.py: Contains the Benchmarker class that runs the matching algorithms and neural networks on different sessions.

Helper files

• MatchingAlgorithms.py: Contains OpenCV image matching algorithms and SuperGlue abstracted into classes to streamline the benchmarking process.
• UniMatch.py: Standardizes datatype for matched keypoints to be used in the image matching algorithms.
• benchmarkers/: Contains different types of Benchmarkers to run in the main.py file.
• mesh_pb2.py: Contains the protobuf class for unpacking the LiDAR data.

Refer to demo.ipynb for a demo of the benchmarkers.