Re:Only Train Once: A One-Shot Neural Network Training and P...
by dheera on Saturday July 17, 2021 @05:25AM
(#27864593)
First of all having millions of images is of itself a good way to combat overfitting.
But yes, you touch upon a very important point, that is the dataset must be sufficiently diverse.
To be fair, humans have this problem as well; if I showed you an octagonal red sign that had the words "GO" inscribed in the middle you may still mistake it for a stop sign at first.
But yes, you touch upon a very important point, that is the dataset must be sufficiently diverse.
To be fair, humans have this problem as well; if I showed you an octagonal red sign that had the words "GO" inscribed in the middle you may still mistake it for a stop sign at first.