Handwriting Recognition?
See original GitHub issueI want to input training data where each character is represented as an array of vectors (I mean each point’s coordinates are given relative to the previous point) with the end
property signifying whether it’s the last point of a stroke.
For example, the letter “T” might look like this:
[ { x: 15, y: 38, end: false}, { x: 49, y: -8, end: false}, { x: 15, y: 4, end: true}, { x: 47, y: 26, end: false}, { x: -4, y: 37, end: false}, { x: 0, y: 4, end: false}, { x: 12, y: 21, end: true} ]
I believe I should use LSTM. But the page says training data must be array of values. Obviously I have an array of hashes. So what do I do?
EDIT: By the way, I have successfully used regular brain.NeuralNetwork
on images (in the form of pixel arrays). But I want to use vectors not pixels.
Issue Analytics
- State:
- Created 4 years ago
- Comments:5
Top GitHub Comments
I see. I am more from ML background a d new to brainjs. If you dont have an actual stroke data it is hard to learn a consistent model for strokes. The typical mnist dataset for example doesn’t provide any stroke data…
On Wed, Oct 30, 2019, 8:10 AM MelerEcckmanLawler notifications@github.com wrote:
Nice!
On Tue, Nov 5, 2019, 2:21 PM MelerEcckmanLawler notifications@github.com wrote: