CoreMon: CoreSet Selection for Training Robust Monkey Trackers in Real-World Environments
The field of object tracking in real-world environments poses unique challenges, especially when tracking agile and unpredictable subjects like monkeys. Successful training of accurate and robust monkey trackers relies on selecting an optimal subset of data from large-scale tracking datasets. This thesis proposal aims to develop CoreMon, an efficient coreset selection approach specifically designed for training monkey trackers in real-world environments.
The primary objective of this research is to design and implement a coreset selection method tailored for training monkey trackers. The proposed approach aims to optimize the data set selection process, ensuring efficient training while maintaining the ability to generalize well in complex real-world scenarios involving monkeys.
Literature Review: Review existing coreset selection techniques and (monkey) tracking algorithms.
CoreMon Design: Develop the CoreMon framework to select an optimal subset of samples in the coreset, considering the trade-off between quality and size.
Implementation and Evaluation: Implement CoreMon in a monkey tracking framework and evaluate its performance on diverse real-world monkey tracking datasets.
Generalization Analysis: Assess the generalization capabilities of monkey trackers trained using CoreMon-selected coresets, reducing biases and overfitting.
CoreMon: An efficient coreset selection approach for training robust monkey trackers in real-world environments.
Comparative Analysis: Demonstrate improved computational efficiency and tracking accuracy of CoreMon compared to existing methods.
Generalization Insights: Provide a comprehensive understanding of CoreMon’s benefits in real-world monkey tracking scenarios.
- Python programming.
- Knowledge of different sampling methods is a plus.
- Data analysis experience is a plus.
- Experience with PyTorch is a plus.