In the actual environment, can the ideal effect be obtained by using realsense camera to estimate the attitude of objects in the data set?

[Xushuangyin]

I used ObjectdatasetTools to make my own datasets, made 10000 pictures of a single object and trained 20 epoch. However, when calling the model to estimate the attitude of the object, the object is fixed and the attitude changes greatly. So I want to use the linemod datasets to train the model and estimate the attitude of the objects in the datasets in the actual environment. Can I get the ideal effect? Where can I buy the objects in the linemod datasets?

The following is my video of using the model to estimate the attitude of objects. You can see that the point cloud model has a wide range of changes.

Have you successfully produced your own data set and achieved the ideal effect? Or does the attitude estimation of objects in linemod data set get the ideal effect? Is it convenient to communicate? thank you!

3d844a26c702f624fea6619a37124476.mp4