Is your feature request related to a problem? Please describe.
Collecting data with the current version of the virtual keyboard is always a bit of a hassle and quite slow because it takes a conscious effort to really hit the keys and keep the hand in proper position due to the lack of haptic feedback. Further, every press that was intended to hit a key but missed it slightly may significantly reduce the quality of the dataset.
Describe the solution you'd like
A simple solution would be to remove the gaps between the keys, if not visually, at least when processing touches. This would enable us to map every position on the canvas to a concrete key, the nearest one. It completely removes the hazard of missing some intentional presses, and should also slightly improve accuracy. But we will most likely have to think of a physical solution to reduce the effort to keep the hand in place.
Describe alternatives you've considered
Another solution might be to switch to a more haptical input method for collecting data, but this would be a suboptimal solution as it introduces a bias for the acceleration data.
Additional context
A mapping of press location to the nearest key (the center of the key) would look like this:
In the picture, each color represents a specific key.
Is your feature request related to a problem? Please describe. Collecting data with the current version of the virtual keyboard is always a bit of a hassle and quite slow because it takes a conscious effort to really hit the keys and keep the hand in proper position due to the lack of haptic feedback. Further, every press that was intended to hit a key but missed it slightly may significantly reduce the quality of the dataset.
Describe the solution you'd like A simple solution would be to remove the gaps between the keys, if not visually, at least when processing touches. This would enable us to map every position on the canvas to a concrete key, the nearest one. It completely removes the hazard of missing some intentional presses, and should also slightly improve accuracy. But we will most likely have to think of a physical solution to reduce the effort to keep the hand in place.
Describe alternatives you've considered Another solution might be to switch to a more haptical input method for collecting data, but this would be a suboptimal solution as it introduces a bias for the acceleration data.
Additional context A mapping of press location to the nearest key (the center of the key) would look like this:
In the picture, each color represents a specific key.