Metric Learning Driven Multi-Task Structured Output Optimization for Robust Keypoint Tracking

Authors: Liming Zhao, Xi Li, Jun Xiao, Fei Wu, Yueting Zhuang

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results have demonstrated the effectiveness of our tracker.
Researcher Affiliation Academia Liming Zhao, Xi Li , Jun Xiao, Fei Wu, Yueting Zhuang College of Computer Science Zhejiang University, Hangzhou 310027, China {zhaoliming, xilizju, junx, wufei, yzhuang}@zju.edu.cn
Pseudocode Yes Algorithm 1: Online Optimization for Tracking
Open Source Code No The paper states they 'create and release a new benchmark video dataset' but does not explicitly mention releasing the source code for their proposed method. It mentions that comparison methods used 'publicly available code' but not for their own work.
Open Datasets Yes We create and release a new benchmark video dataset containing four challenging video sequences (covering several complicated scenarios) for experimental evaluations.
Dataset Splits No The paper mentions using video sequences for experiments and refers to 'training samples collected from the results of previous K frames' for model updates, but it does not specify explicit train/validation/test dataset splits (e.g., percentages or counts for each partition).
Hardware Specification Yes On average, our algorithm takes 0.0746 second to process one frame with a quad-core 2.4GHz Intel Xeon E5-2609 CPU and 16GB memory.
Software Dependencies No The paper states: 'We also implement our approach in C++ and OPENCV.' However, it does not specify version numbers for C++ or OPENCV, or any other software dependencies.
Experiment Setup Yes For metric learning, the linear transformation matrix M is initialized to be an identity matrix. For multi-task learning, the number of tasks K is chosen as 5 and we update all the multi-task models frame by frame. All weighting parameters λ1, λ2, ν1, ν2 are set to 1, and remain fixed throughout all the experiments.