This code combines Circulant Matrix Tracker by João F. Henriques (Github, Author's page) and Hardnet image descriptor (Github, Paper)
- Get datasets. Comes with loader for MILTrack and VOT. Alternatively you can write your own loader.
- Extract somewhere. Recommended structure is to put data into
./data/sets/*
and create folder./data/logs/
where output will be generated - Run
./circulant_matrix_tracker.py -i path_to_dataset
For more options run ./circulant_matrix_tracker.py -h
To change image descriptor to be used for tracking use -d|--descriptor
option. Currently supported descriptors are:
raw
|gray
- Describe image using raw grayscale pixelshardnet
- Describe image using 128-channel pretrained HardNet++ descriptor
- WIP: GPU computation. If enabled some things get offloaded to gpu, but this parts needs more optimizations work
- TODO: HOG Features descriptor
- TODO: Some metric to automatically measure and evaluate quality of tracking
- TODO: Rewrite everything to c++
- Python (used 2.7, might work with other versions too)
- Numpy
- Matplotlib
- Scipy
- Torch
- PyLab