A pose estimation method combining instance segmentation
I have implemented a 6D pose estimation method for objects based on RGBD information, which combines instance segmentation. (1) Obtain object masks through instance segmentation; (2) Crop to obtain RGB and depth information of the area where the object is located; (3) Extract RGB and depth information features using convolutional networks, and then perform feature fusion; (4) Calculate the loss by combining the feature vectors with the features of the model; (5) Regress the translation and rotation of the object relative to the camera coordinate system.
This project is extended base on