Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
kingeorge 1063ffa405 | 1 year ago | |
---|---|---|
src | 1 year ago | |
README.md | 1 year ago | |
test.py | 1 year ago | |
train.py | 1 year ago |
OmniPose是一个人体姿态估计网络,它结合了多尺度的特征 multi-scale features,并引入了一个提升过的 waterfall 结构,可以在保持特征图高分辨率的情况下增大感受野FOV,这个提升过后的 waterfall 结构称为 waterfall Astrous Spatial pooling(WASPv2),它既扮演特征提取器,也扮演 decoder;通过引入高斯热图调制(Gaussian heatmap modulation)方法来帮助进行 joint 坐标点的定位,可以使 upsampling 过程中的 point 定位的更加准确,克服了分辨率上升而存在的量化误差(quantization error)问题。
使用的数据集:COCO2017
└──OmniPose
├── README.md
├── src
├── utils
├── coco.py # COCO数据集评估结果
├── nms.py # nms
├── transforms.py # 图像处理转换
├── config.py # 参数配置
├── dataset.py # 数据预处理
├── network_with_loss.py # 损失函数定义
├── omnipose.py # 主干网络定义
├── wasp.py # wasp模块定义
└── predict.py # 热图关键点预测
├── eval.py # 评估网络
└── train.py # 训练网络
配置请参考脚本config.py
。 通过官方网站安装MindSpore后,您可以按照如下步骤进行训练和评估:
参数 | Ascend |
---|---|
模型 | Omnipose |
资源 | Ascend 910 |
MindSpore版本 | 1.5.1 |
数据集 | COCO |
训练参数 | epoch=210 |
输出 | AP |
准确率 | AP 0.7953 |
2022昇腾AI创新大赛昇思赛道 第二批 赛题十一:利用MindSpore实现OmniPose姿态估计网络
Python
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》