Fractal Video Dataset
收藏资源简介:
Fractal Video Dataset是由国立雅典技术大学电气与计算机工程学院创建的,用于动作识别任务的预训练。该数据集通过分形几何自动生成大量短的合成视频片段,具有显著的多样性,源于分形生成复杂多尺度结构的能力。数据集的创建过程包括生成分形视频、模拟人类动作的增强以及训练过程。该数据集主要应用于动作识别领域,旨在解决从视频或传感器数据中准确检测和解释人类动作的问题,广泛应用于监控、医疗、机器人、体育分析和人与计算机交互等领域。
The Fractal Video Dataset was developed by the School of Electrical and Computer Engineering, National Technical University of Athens for pre-training in action recognition tasks. This dataset automatically generates a large volume of short synthetic video clips through fractal geometry, boasting remarkable diversity stemming from the capacity of fractals to produce complex multi-scale structures. The dataset construction process encompasses three core parts: generating fractal videos, performing data augmentation that simulates human motions, and finalizing the training workflow. Primarily utilized in the field of action recognition, this dataset is designed to address the challenge of accurately detecting and interpreting human actions from video or sensor data, and has been widely applied in domains including surveillance, healthcare, robotics, sports analytics, and human-computer interaction.

- 1Pre-training for Action Recognition with Automatically Generated Fractal Datasets国立雅典技术大学电气与计算机工程学院 · 2024年



