Countix-AV

Countix-AV 是一个数据集，用于通过重新利用 Countix 数据集创建的视觉和声音重复动作计数。它是通过从 Countix 中选择 19 个类别创建的，这些类别的重复动作具有清晰的声音，例如鼓掌、打网球等。数据集包含 1,863 个视频，其中 987、311 和 565 个用于训练、验证和测试。作者维护了来自 Countix 的原始计数注释，并为每个视频保留了相同的拆分（即训练、验证或测试）。

Countix-AV is a dataset designed for visual and auditory repetitive action counting, which is developed by repurposing the Countix dataset. It is constructed by selecting 19 categories from Countix where their repetitive actions have distinct auditory signatures, such as clapping, playing tennis, and so on. The dataset comprises 1,863 videos in total, with 987, 311, and 565 samples assigned to the training, validation, and test sets respectively. The authors retain the original counting annotations from Countix and preserve the identical data split (i.e., training, validation, or testing) for each video.

搜集汇总

背景与挑战

背景概述

Countix-AV是一个用于视觉和声音重复动作计数的数据集，包含1,863个视频，覆盖19个类别，分为训练、验证和测试集。该数据集由阿姆斯特丹大学和阿联酋起源人工智能研究院于2021年发布，适用于计算机视觉和声音分析研究。