I set out to get my Kinect camera feed working inside Unity and turned to Gemini for help. It wasn’t a very smooth process, but I did get to try out Roo Code finally and see first-hand how expensive some of these new LLMs are.
Once I got the texture flipped (and also discovered a simpler, alternative method using a WebcamTexture), I then set up a separate camera to render my skeleton and particles as an overlay.
Scaling the skeleton to the right size took some tinkering, and after some head scratching, I realized that the reason my particles sometimes disappeared while scaling was the maxDistanceFromCamera flag I’d set up in my scene configuration.
I quickly got sidetracked and decided to investigate body segmentation because I’d been thinking about it the past few days. I found a video from Keijiro on his Bodypix library that showed the segmentation in action. I was stoked because the repo had just recently been updated to work in Unity 6. However, when I looked inside his project, I couldn’t find the same sample scene from the video with the segmentation and face pixelization.
I decided that sample must have been archived with the old version and used Gemini to recreate it. It took way longer than it should have because of constant user error on my end.
Still, I did get it working eventually. That felt good. And then I rewatched the video and realized the repo Keijiro used for it was still on his GitHub, and was still functional. Cheers to reinventing the wheel 🎡