Open weibao1001 opened 2 years ago
Since our estimation is a 2d -> 3d lifting method, we need to input 2d pose along with the point cloud features into the 3d regression module. In our paper, we use Hourglass network to estimate 2d pose, while the released code omits the 2d estimation and replaces it with the 2d ground truth for simplification
由于我们的估计是 2d -> 3D 提升方法,因此我们需要将 2d 姿势以及点云特征输入到 3d 回归模块中。在我们的论文中,我们使用沙漏网络来估计2d姿势,而发布的代码省略了2d估计并将其替换为2d地面实况以进行简化。
Excuse me, does the test speed of 50fps mentioned in the paper mean 2d -> 3D or 50fps calculated from the input data to the last?
There are Labels in the test process of the ITOP dataset. Why does the image_coordinate affect the test results?