Can I traning additional models?
#2
by
hyeonbok
- opened
Thank you for making your great model. I have created a program that automatically inputs XMP metadata by applying your model. The program works well and accurately recognizes character information. However, I will continue to encounter new characters in the future. Is there a way to traning additional models?
Given you have enough samples per character, the absolutely simplest way would be to:
- extract the features for those characters using the layer immediately before the final Dense layer
- train a simple linear classifier (a Keras Dense layer, or equivalent in your frameworks of choice) on the extracted features.
Afterwards you could merge your own classifier with the original Dense layer, or add it as an additional output head.