Action with RAre Scene (ARAS)
收藏资源简介:
ARAS数据集是由香港中文大学的研究团队创建,专门用于评估动作识别模型在罕见场景下的表现。该数据集包含1038个视频片段,每个片段都与Kinetics-400中的特定动作类别相关联,但场景设置为罕见或不常见。创建过程中,研究团队通过在YouTube上使用特定动作和罕见场景的组合作为查询,筛选并手动检查视频以确保其符合要求。ARAS数据集的应用领域主要集中在动作识别技术中,特别是在模型需要处理多样化和非典型场景时的性能评估。
The ARAS dataset was developed by a research team from The Chinese University of Hong Kong, specifically designed to evaluate the performance of action recognition models in rare scenarios. This dataset comprises 1038 video clips, each linked to a specific action category within Kinetics-400, but with rare or atypical scene setups. During its creation, the research team utilized combinations of target actions and rare scenarios as search queries on YouTube, before screening and manually verifying the videos to confirm their compliance with the dataset requirements. The ARAS dataset is primarily applied in the field of action recognition, especially for assessing model performance when handling diverse and non-standard scenarios.

- 1Mitigating Representation Bias in Action Recognition: Algorithms and Benchmarks香港中文大学 · 2022年



