Evaluating Trained Models

[peterdavidfagan]

Hi there,

Thank you for open-sourcing this repository. I am in the process of evaluating the trained models on my own image data so I can include them in a ROS 2 grasping package. I am currently struggling to get the trained models to work well. I am evaluating the following model, below are my initial qualitative results.

Note: RGB image is actually BGR below but this has been fixed in latest code and the results remain the same.

I plan to update my iterations on this in the following issue including sharing code.

Current Thoughts:

• [ ] Is the model overfitted to the camera pose/s in the Cornell Grasping Dataset?