metadata
license: cc-by-4.0
tags:
- basketball
- classification
Title: Basketball-Related Objects Classification Dataset
Relation to Paper: Developed as an additional sub-task to demonstrate how the precise detection of basketball-related objects from E-BARD can be leveraged for domain-specific classification, which significantly improved the performance of the fine-tuned EBQwen 3B model.
Dataset Statistics:
- Consists of 22,210 human-annotated basketball-related entities.
- Objects are categorized into predefined classes: basketball, hoop, player, and referee.
- Serves as a crucial fine-tuning dataset to help vision-language models achieve high F1-score performance across all specific domain classes.