brilliantsong2/RFL

Code for "Recurrent Filter Learning for Visual Tracking"

Recurrent Filter Learning for Visual Tracking

This is the implementation of our RFL tracker published in ICCV2017 workshop on VOT. Our code is written in python3(3.5) using Tensorflow(>=1.2) toolbox

For easy comparison, we upload our OTB100 results files to the main directory ./otb100_results.zip

Tracking

You use our pretrained model to test our tracker first.

1. Download the model from the link: https://drive.google.com/open?id=0BzxOz7xyra_-dzJaY2d0Y1RiZFk
2. Put the model into directory ./output/models
3. Run python3 tracking_demo.py in directory ./tracking

Training

1. Download the ILSRVC data from the official website and set proper paths for ISLVRC and their tfrecords in config.py
2. Then run the process_data.sh in ./data_preprocssing directory to convert ILSVRC data to tfrecords.
3. Run python3 train.py to train the model.

If you find the code is helpful, please cite

@inproceedings{Yang2017,
    author = {Yang, Tianyu and Chan, Antoni B.},
    booktitle = {ICCV Workshop on VOT},
    title = {Recurrent Filter Learning for Visual Tracking},
    url = {http://arxiv.org/abs/1708.03874},
    year = {2017}
}