This is a fork of the Python implmentation of KCF. I have modified the code to run sequences from the VOT Challenge and provide Accuracy, Robustness and Frames Per Second evaluation metrics.
Example from VOT2016/matrix sequence:
The original paper:
High-Speed Tracking with Kernelized Correlation Filters
J. F. Henriques, R. Caseiro, P. Martins, J. Batista
TPAMI 2015
- Python 2.7
- NumPy
- Numba
- OpenCV (ensure that you can
import cv2
in python)
I recommend installing Anaconda(for Python 2.7), and OpenCV 3.1(from opencv.org).
- Install the required packages
git clone https://github.com/uoip/KCFpy.git
cd KCFpy
- Download a dataset from the VOT Challenge. webpage. eg. VOT2016 dataset
- copy the dataset into the 'VOTdataset' folder in KCFpy
- from the command window, call run.py with 1 paramater: The path to the sequence you wish to run from the 'VOTdataset' folder
eg:
python run.py VOT2016/matrix
The figures will automatically be saved into the 'saved' folder. Comment out line 192 in run.py to stop this.
I used Anaconda for Python 3 and setup a python 2.7 virtual environment called 'pykcf' on Windows 10.
How I did this:
conda create --name pykcf python=2.7.13 numpy numba scipy
activate pykcf
pip install openCV-python