Closed JinhangZhu closed 4 years ago
利用了Left/Right Hand Segmentation in Egocentric Videos论文中的经验模型,以及其他传统计算机视觉知识判断左右手,见https://github.com/JinhangZhu/project-diary/blob/master/differentiate.ipynb
第一篇paper:用的是基于SVM的segmentation+基于Maxwell function和经验模型的判别方法,我自己stage-2的classifier-b有参考,初稿中也做了比较;
第二篇paper:论文是将无修改的YOLOv3模型运用于多个手部数据集,做了一个没有很多创新的贡献,左右手识别基于相对位置,我参考了其中的论文写作。
第三篇paper:觉得如果用在论文里,有点太像拼凑tricks了,只看了博客的总结,没有读论文。实际这篇博客提到的论文采用的解决occlusion的方法,我用别的小tricks代替了,菜是原罪。
- [x] 扩展数据集的过程中踩坑很多,之后总结
Plan
[x] 实现设计的prototypes
[x] 扩展数据集(包含上传supervisely手动纠正)
[x] 对模型用大数据集彻底地训练
看
Left/Right Hand Segmentation in Egocentric Videos Paper
Egocentric Hand Track and Object-based Human Action Recognition Paper
遮挡问题相关论文 Post1