Back on the train and in the glasses deck. Sitting next to a big coffee spill on the flor and a man slowly, methodically eating a sandwich.

I don't know if I'll actually publish this as a post, but I wanted to collect my thoughts by writing.

This morning I was thinking of buying a new kitchen knife and working on cooking a couple of dinners a week again. I'd like to find some sort of system for all of that that feels good.

I was also thinking about buying some new pants too.

Is that today's todo list?

• Buy kitchen knife (best starter knife?)
• Buy pants

I got the little raspberry pi TV working. That's cool I think. Still somehow feel behind on the work I'm doing.

I want to take some time to redo my Neovim config:

• add config to navigate wrapped lines
• remove ChatGPT integration
• add neogit
• add diffview I think

What do I want to do for a constraint systems thing? I was still thinking about manipulation of digital objects in the real world. Just keying out a color didn't seem to work that well. I still feel like that should work, maybe I just need white paper background and better lighting... but hands would still disrupt something like red. Maybe that's why it's a green and blue screen, red has more collisions with skin tones...

I think I figured out dynamicland is using SIFT to track objects. Which should handle rotation and different sizing. I think I could get that going. I still don't quite understand how you choose what features. Maybe you feed it an image of just the paper with the markers on it. I wonder why they're called fiduciary markers.

If I got that going I think the best thing to do would be to let the user put together a digital canvas but with physical manipulation.

I'm hot on this train.

I was trying to think if I could use lightweight object detection model instead - maybe if I printed out really generic objects, like cats and dogs... but that wouldn't necessarily handle rotation.

I also know I have good hand landmark tracking - could I use that instead? I feel like the gap there is a good selection mechanism. It could be fun to do forefinger pinch and drag to move. Middle finger pinch and drag to scale. Ring finger pinch to open context menu. Pinky finger pinch to delete. Set a little timer on it. Ah what about moving camera though. Other hand? Yea other hand pan and zoom.

Set up canvas. Could replicate with mouse and keyboard keys.

Camera video goes behind the canvas. Maybe you could pull in resources off the side? Textboxes could work too.

Oh do you want to think more about how your zoom canvas should work? I don't understand why the center doesn't go to zero zero in the current canvas. It's kind of OK but it's not totally on stable ground.

What was the thing I wanted to write about, adding computation to objects.

dynamicland - bless and draw four points - that was actually the older model right?

hyperphotos - clickable photos. still a gemini demo where I could just use bounding boxes for that. I need a source for the links. Something like a sitemap, or just a list of links. Hmm maybe a site builder. Where you first specify the pages, and it just does a bare-canvas. But you have the option to then go to a webcam view.

What's a good bare-bones structure? Portfolio and about? If it was hamster photos and you put text descriptions... Still wouldn't handle rotation. Match these images to this sketch...