Provide Isaac SDK a tool for 6D Pose Estimation of Cardboard Boxes?

[El_HH]

Hello Isaac-family,

I recently started a project in the field of 6D Pose Estimation of multiple cardboard boxes in shipping containers. The aim is to detect as many cardboard boxes as possible in the container. Then to estimate the 6D Pose of each detected box. Afterwards the information of the poses are used as input for a autonomous unloading system.

Currently I’m reading a lot of papers and most of them claim to use synthetic data for training/testing. Furthermore they are based on a preknown 3D Model of the object of interest. But in real world, cardboard boxes has not only one sizes as the 3D Model. I also saw a tutorial for 6D Pose Estimation for pallets and boxes, but they are standardized. Does Isaac provide a solution for 6D Pose Estimation of multiple different sized cardboard boxes?

I put here some Images as example of the cardboard set up.

[hemals]

You intend to perform this pose estimation using a monocular camera or a stereo pair (with or without structured light)? Do you know the sets of box dimensions you will encounter beforehand? It is difficult to determine the depth of a box from a flat frontal image until you start removing layers of boxes.

You can train models to determine the bounding areas of different boxes as you have in the first image as an image segmentation task. From there, you really have no further information to work with until you manipulate the box or match the width/height into a unique depth value.