Single object tracking, implemented from SINT

• mxnet: 0.9.5
• numpy
• scipy

• Beijing Nas: '/data01/SOT/models'
• This folder includes all the models.

• ALOV++ dataset is used for training, the seqs and annotations are located at '/data01/SOT/datasets/ALOV'
• OTB dataset is used for evaluation, the seqs and annotations are located at '/data01/SOT/datasets/OTB'

• Refer more details to 'train/train_SINT_triplet.py'
• example usage

python train_SINT_triplet.py --lmnn --rand_mirror --fine_tune --lr 0.001 --lmnn_threshd 0.9 --gpus 0

• The 'test/' folder is used to test the trained model on OTB dataset. The tracking results will be outputted to a directory.
• example usage

python eval_SINT_train.py --overlapthresh 0.6 --topK 5 --numangles 20

• The 'evaluation/' folder is used to compare tracking results with other trackers. More details can be found from Visual Tracker Benchmark
• example usage

src_root = '/dir/to/your/generated/tracking_results'
prepare_for_evaluation(src_root)

Then add your tracker to '/evaluation/util/configTrackers.m'

main_running.m % generate figures in '/evaluation/figs/overall' folder
main_seperate_plot.m % used for debug, will generate figures in '/evaluation/figs/overall_seperate' folder