D
Deleted member 36445
Guest
I've been thinking these days about an idea: training an AI model with videos of cube solving and their reconstructions.
It could use a real smart cube and the JSON data from it. Also, it would be good if there were two separate filmings, one from a different angle, for more sampling.
Then, we would feed an AI model with the filmings and JSON data, and we could get the same JSON from any filming, even if the cube is not a smart cube.
One more thing:
I had another idea, using a specially stickered cube for better image recognition, like black and white stickers, or other configurations, that would allow a program to recognize the movements more efficiently.
It's unfortunate that there aren't enough good cube recognition software programs at the moment, despite the vast amount of technology we have available.
It should be possible to have a better way of capturing cube states from photos or videos, not just 3x3, but even other puzzles.
It could use a real smart cube and the JSON data from it. Also, it would be good if there were two separate filmings, one from a different angle, for more sampling.
Then, we would feed an AI model with the filmings and JSON data, and we could get the same JSON from any filming, even if the cube is not a smart cube.
One more thing:
I had another idea, using a specially stickered cube for better image recognition, like black and white stickers, or other configurations, that would allow a program to recognize the movements more efficiently.
It's unfortunate that there aren't enough good cube recognition software programs at the moment, despite the vast amount of technology we have available.
It should be possible to have a better way of capturing cube states from photos or videos, not just 3x3, but even other puzzles.
