Build a dataset using only rgb images and aligned 16-bit depth images?

[Yuan-TS]

Hello, does this tool require a 3D model? I only have RGB images and aligned 16-bit depth images. Can I use this method? If not, is there any way to build a dataset using only the above two data?

[AnasIbrahim]

Hi, This tool requires a 3D model. So you cannot use it in a straightforward way.

But, you actually can have 6D Pose annotations without a 3D Model. If and only if, you are interested in 6D Pose annotations and no 2D segmentation masks. Simply, you can use a bounding box. So, you can change this tool to use a bounding box instead of a 3D model.

If you answer these questions I can help you more:

• Is the object static and the camera is moving? or the camera moves and the object is static?
• if the camera is moving can it be somehow tracked?
• Are the images taken in a sequence from a scene?
• How big is your data? are you making a small evaluation data or collecting big training data?

I'm already updating this tool to annotate sequences when the camera is moving and the objects are static. So I committed this to the BOP toolkit but I currently still have these latest updates not pushed yet my own fork.