Generation of clock faces for a clock face reader
There are many potential models including
Hour 0-11 = 12 Categories Minutes = Single output for regression
One of the challenges is the step change from 11->0 and 59->0
Hour 0-11 = 12 Single output for regression Minutes = Single output for regression
One of the challenges is the step change from 11->0 and 59->0
2 outputs per parameter
Hour sin and cos Minutes sin and cos
The hour can be the digit hour ( 0, 1, 2, etc...) or it can be the analogue hour. An analogue displays the hour in a form such as 12.5. The minute is encoded in the hour hand angle.
Creating an accurate model, with perfect data is relatively simple. However creating a robust model, which can is robust to different clock faces, real time images from a web cam is a little different.
There are three scripts for generating clock faces.
Matplotlib is used to generate clockface images
Tensors are used to generate clock faces.
Tensors and a clockface png or jpg image are used to generate clock faces. You must provide this clockface image.
It iss quite old and needs updating. A companion csv file is also generated. This provides labels for training a neural network.
A better approach might be to generate everything on the fy during training, this includes clock face, hands etc...
The labels are filename, hour hand angle ( cos, sin ), minute hand angle ( cos, sin)
Rather than using angles, the sin and cos each angle have been taken, so there are two parameters per handle angle ( The aim was to avoid step changes in the value )
This code from a few years ago, just for parctice purposes, so I am currently reviewing it