T-CNN

pytorch implement of "Modeling and propagating cnns in a tree structure for visual tracking"

original paper:

@article{nam2016modeling,
  title={Modeling and propagating cnns in a tree structure for visual tracking},
  author={Nam, Hyeonseob and Baek, Mooyeol and Han, Bohyung},
  journal={arXiv preprint arXiv:1608.07242},
  year={2016}
}

requirements

• python 3.6+
• pytorch
• torchvision
• scikit-learn

execute

usage: main.py [-h] dataset dest

positional arguments:
  dataset
  dest

this code use VOT dataset and will use ground truth of first frame as tracking target

note

in original it use VGG-M for feature extraction. Since there's no pretrained VGG-M model in PyTorch, it use vgg11_bn in this code and the output image size could be same in the paper.