Any thoughts on segmentation?

[Michael Wimble]

In the coming months I’m hoping to turn on segmentation for Sigyn. Oooh, I like the alliteration. I think I want to use it to:

• Discover power cords, floor mats an similar clutter without attempting to do so using depth maps.
• Discover the floor, table tops, walls — flat surfaces that will help me in my project to do optical localization without relying on LIDAR and odometry.
• Discover power outlets, sliding door handles, refrigerator handles and other objects I need to interact with.

I’m not sure if simple segmentation will do the trick or if I need semantic segmentation.

I also haven’t tried it in years, especially I don’t remember if I tried it with the OAK-D camera.

Has anyone played with them? Has anyone implemented them?

Pointers and thoughts would be helpful.

[Alan Timm]

Now that you mention it Jim put a couple of projects on our radar last weekend that were potentially very interesting.

real time object detection

https://github.com/NVIDIA-AI-IOT/nanoowl

real time object segmentation

https://github.com/NVIDIA-AI-IOT/nanosam