Improving the NVIDIA Isaac 3D Pose Estimation Model with Feature Map Explorer

Originally published at:

As many would probably agree, the development of a winning deep learning model is an iterative process of fine-tuning both the model architecture and hyperparameters. The tuning is often based on insights about the network behavior gained by examining various data obtained in training. One such data is the output from each layer of a…



It is mentioned in the post that evaluation metrics have been run on the simulated data for the dolly object. Is there any way I can implement the same for custom objects? Where can I find the code used for evaluation?

It will be part of the next Isaac SDK release which is towards the end of next month.