Home

Awesome

LADCF - No 1 Algorithm on the public dataset of VOT2018

Demo for Learning Adaptive Discriminative Correlation Filters (LADCF) via Temporal Consistency Preserving Spatial Feature Selection for Robust Visual Object Tracking

Download the Paper

@article{xu2019learning, title={Learning Adaptive Discriminative Correlation Filters via Temporal Consistency Preserving Spatial Feature Selection for Robust Visual Object Tracking}, author={Xu, Tianyang and Feng, Zhen-Hua and Wu, Xiao-Jun and Kittler, Josef}, journal={IEEE Transactions on Image Processing}, pages={5596--5609}, volume={28}, number={11}, year={2019} }

The tracker codes for ICCV2019 can be download here.

More group feature selection strategies are explored.

The tracker codes for VOT2018 can be download here.

More powerful features and data augmentation techniques are added for the VOT2018.

Instruction for LADCF_HC Tracker:

Learning Adaptive Discriminative Correlation Filter on Low-dimensional Manifold (LADCF) utilises adaptive spatial regularizer to train low-dimensional discriminative correlation filters. We follow a single-frame learning and updating strategy: the filters are learned after tracking stage and then updated using a fixed rate [1]. We use HOG [2] and CN [3]. Code modules refer to ECO [4] in feature extraction.

Dependencies:

Operating system:

Ubuntu 14.04 LTS, Matlab R2016a, CPU Intel(R) Xeon(R) E5-2643

References:

Raw Results:

OTB100(hand-crafted feature) OTB100(deep feature)