I want to create a model for classifying single channel images that does not look like at all to imagenet data.
I already have done it with Keras+TF and now I want to do it with fastai and compare the accuracy, time taken, … .
How can I define layers like Convolution, BN, Maxpooling, FC and etc in fastai. Can you point me to a helfull example notebook.
I am in the third lesson and so far every thing is based on using an existing imagenet model.
Take a look at the torchvision’s canned models implementations.
I’m curious. What do your images look like?
They are B/W drawings from various technical fields.