IPN Hand
收藏资源简介:
IPN Hand数据集是由国立电讯大学和墨西哥国立理工学院共同创建,专注于连续手势识别,包含超过4000个手势样本和800,000帧RGB视频,涉及50名不同参与者。数据集设计了13种静态和动态手势,主要用于无触摸屏幕交互。创建过程中,数据收集自约30个不同场景,考虑了背景和光照的实际变化。该数据集的应用领域广泛,包括汽车行业、消费电子、家庭自动化等,旨在解决连续手势识别中的实时性和准确性问题。
IPN Hand Dataset was jointly developed by the National Telecommunications University and the National Polytechnic Institute of Mexico (IPN). Focused on continuous gesture recognition, this dataset comprises more than 4,000 gesture samples and 800,000 RGB video frames, involving 50 distinct participants. The dataset includes 13 categories of static and dynamic gestures, which are primarily designed for touchless screen interaction. During the creation of this dataset, data was collected from approximately 30 distinct scenarios, with practical variations in background and lighting taken into full consideration. This dataset finds broad application across multiple domains including the automotive industry, consumer electronics, home automation and more, aiming to address the challenges of real-time performance and accuracy in continuous gesture recognition.
IPN Hand 数据集概述
数据集详情
视频规格
- 分辨率:640x480
- 帧率:30 fps
- 视频类型:仅包含 RGB 视频
手势定义
- 总手势数:21 种
- 控制指针和无触摸屏幕交互的手势:13 种
手势统计
| id | 标签 | 手势 | 实例数 | 平均持续时间(标准差) |
|---|---|---|---|---|
| 1 | D0X | 非手势 | 1431 | 147 (133) |
| 2 | B0A | 单指指向 | 1010 | 219 (67) |
| 3 | B0B | 双指指向 | 1007 | 224 (69) |
| 4 | G01 | 单指点击 | 200 | 56 (29) |
| 5 | G02 | 双指点击 | 200 | 60 (43) |
| 6 | G03 | 向上抛 | 200 | 62 (25) |
| 7 | G04 | 向下抛 | 201 | 65 (28) |
| 8 | G05 | 向左抛 | 200 | 66 (27) |
| 9 | G06 | 向右抛 | 200 | 64 (28) |
| 10 | G07 | 双开 | 200 | 76 (31) |
| 11 | G08 | 单指双击 | 200 | 68 (28) |
| 12 | G09 | 双指双击 | 200 | 70 (30) |
| 13 | G10 | 放大 | 200 | 65 (29) |
| 14 | G11 | 缩小 | 200 | 64 (28) |
| 所有非手势 | 1431 | 147 (133) | ||
| 所有手势 | 4218 | 140 (94) | ||
| 总计 | 5649 | 142 (105) |
基准结果
- 基准结果:可在 此处 找到 IPN Hand 数据集的孤立和连续手势识别的基准结果。
使用要求
- 软件要求:
- Python 3.5+
- PyTorch 1.0+
- TorchVision
- Pillow
- OpenCV
预训练模型
- 模型列表:
- ResNeXt-101 模型
- ResNet-50 模型
- HarDNet 模型(即将发布)
- 光流模型
使用方法
准备
-
从 此处 下载数据集。
-
克隆此仓库: console $ git clone https://github.com/GibranBenitez/IPN-hand
-
将所有预训练模型存储在
./report_ipn/目录下。
孤立测试
- 修改
./tests/run_offline_ipn_Clf.sh中的数据集路径并运行: bash $ bash run_offline_ipn_Clf.sh
连续测试
- 修改
./tests/run_online_ipnTest.sh中的数据集路径并运行: bash $ bash run_online_ipnTest.sh
引用
如果您在研究中使用了 IPN Hand 数据集,请引用以下论文:
bibtex @inproceedings{bega2020IPNhand, title={IPN Hand: A Video Dataset and Benchmark for Real-Time Continuous Hand Gesture Recognition}, author={Benitez-Garcia, Gibran and Olivares-Mercado, Jesus and Sanchez-Perez, Gabriel and Yanai, Keiji}, booktitle={25th International Conference on Pattern Recognition, {ICPR 2020}, Milan, Italy, Jan 10--15, 2021}, pages={1--8}, year={2021}, organization={IEEE}, }
许可证
- 代码许可证:MIT 许可证
- 数据和标注许可证:Creative Commons Attribution 4.0 License

- 1IPN Hand: A Video Dataset and Benchmark for Real-Time Continuous Hand Gesture Recognition国立电讯大学 · 2020年



