Related Experiment Video
Updated: Jan 2, 2026
05:57
Long-term Video Tracking of Cohoused Aquatic Animals: A Case Study of the Daily Locomotor Activity of the Norway Lobster Nephrops norvegicus
Published on: April 8, 2019
7.2K
GOT-10k: A Large High-Diversity Benchmark for Generic Object Tracking in the Wild
IEEE Transactions on Pattern Analysis and Machine Intelligence
|December 6, 2019
Summary
We introduce GOT-10k, a large dataset for training and evaluating object tracking algorithms. This diverse dataset features over 10,000 videos and promotes class-agnostic tracker development.
Area of Science:
- Computer Vision
- Machine Learning
Background:
- Object tracking is crucial for various applications.
- Existing datasets lack diversity and comprehensive coverage of moving objects.
- There is a need for a unified platform for training and evaluating trackers.
Purpose of the Study:
- Introduce GOT-10k, a large-scale, high-diversity dataset for object tracking.
- Provide a unified training and evaluation platform for class-agnostic, generic short-term trackers.
- Promote the development of more robust and generalizable tracking algorithms.
Main Methods:
- Constructed GOT-10k using WordNet's semantic hierarchy for comprehensive class population.
- Included over 10,000 video segments with more than 1.5 million labeled bounding boxes.
- Introduced a one-shot evaluation protocol with zero-overlap between training and testing classes.
Main Results:
- GOT-10k covers over 560 object classes and 87 motion patterns, significantly wider than existing datasets.
- The dataset enables unified training and stable evaluation of deep trackers.
- Extensive experiments with 39 tracking algorithms were conducted on GOT-10k.
Conclusions:
- GOT-10k provides a comprehensive and unbiased resource for advancing object tracking research.
- The one-shot protocol encourages the development of trackers that generalize to unseen objects.
- A community platform with evaluation toolkits and leaderboards is established to facilitate research.
