SiamMCF

Implementation in Python and Tensorflow of the tracker described in the paper Multiple context features in siamese networks for visual object tracking, VOT 2018 - ECCVW.

Note 1: The code supports both Python 2 and 3. However, the results vary depending on the Python version. We used Python 2.7 to generate the results shown in the paper.

Note 2: The instructions in this guide are for bash in Linux. If you use another environment, you may have to adapt them accordingly.

Requirements

• Python 2 or 3 (Tested on Python 2.7 and 3.6)
• Tensorflow (Tested on tensofflow 1.4)
• Matplotlib
• Numpy
• Pillow
• Scipy

Installation

1. If you do not have the requirements above, install them first, following the instructions provided by each package.

2. Clone this repository:

   git clone https://github.com/hmorimitsu/siam-mcf.git

   We will refer to the directory generated by the clone operation as siam-mcf. If you cloned the repository with another name, be sure to adapt the following instructions accordingly.

3. Download the trained checkpoint from Google Drive or Baidu Yun and extract them to siam-mcf/pretrained. Your directories should look like the following:

   + siam-mcf/
     + pretrained/
       - siam_mcf.ckpt-50000.data-00000-of-00001
       - siam_mcf.ckpt-50000.index
       - siam_mcf.ckpt-50000.meta
   

4. Add a (symlink to a) dataset in VOT or OTB style in siam-mcf/data. For example, supposing you added the vot16 dataset, your directories should look like the following:

   + siam-mcf/
     + data/
       + vot16/
         + bag/
         + ball1/
         ...
   

5. Run the evaluation script:

   cd siam-mcf
   python run_tracker_evaluation.py --dataset_name vot16 --save_bboxes --save_images

   The image and bounding box results will be saved in the directory siam-mcf/results. You can see the command line parameters accepted by the script by running:

   python run_tracker_evaluation.py -h

   Additional parameters can be changed in the JSON files in siam-mcf/parameters.

Raw results

The raw results obtained in OTB13, OTB15, VOT16 and VOT17 datasets can be downloaded from Google Drive or Baidu Yun.

References

If you find this work useful, please consider citing:

@inproceedings{morimitsu2018multiple,
  title={Multiple context features in siamese networks for visual object tracking},
  author={Morimitsu, Henrique},
  booktitle={ECCV 2018 Workshops},
  year={2018}
}

@inproceedings{bertinetto2016fully,
  title={Fully-convolutional siamese networks for object tracking},
  author={Bertinetto, Luca and Valmadre, Jack and Henriques, Jo{\~a}o F and Vedaldi, Andrea and Torr, Philip H S},
  booktitle={ECCV 2016 Workshops},
  pages={850--865},
  year={2016}
}

License

This software includes third party open source software components from:

• SiamFC: https://github.com/torrvision/siamfc-tf,
• VOT-toolkit: https://github.com/votchallenge/vot-toolkit.

Original code for SiamMCF is licensed under the BSD-3-Clause license (LICENSE_siam_mcf.md). Code modified from SiamFC may also be subject to their license (LICENSE_siam_fc.md). Code from VOT toolkit in src/vot is licensed under the GPLv3 license(LICENSE_vot.md).

Contact

If you have any questions or find any bugs in the code, please feel free to contact me at [email protected].