Is it possible to make custom loss function using opencv + numpy?

autograd
1

Hi,

I want to create a custom loss function to train the model.

What the model is learning is a 2d transformation matrix.

To do this, the feature points of the image are detected and matched. (using opencv + numpy)

As a result,

loss = (the distance between feature points).sum() / num_matche

I want to train the model to make loss zero.

is it possible to make custom loss function using opencv + numpy?

If so, is there an example that implements this?

2

Yes, have a look at this numpy extensions tutorial.