Lesson 2 In-Class Discussion ✅

[suvash (Suvash Thapaliya)]

This might be a slightly advanced thing, better to ask there, also there’s some discussion spread around here.

How to get an empty ConvLearner for single image prediction?

Is there a way to instantiate a ConvLearner without specifying a DataBunch or having to download the pre-trained backbone, into which we can simply load previously saved weights into? For example, I’m imaging something like this is possible … learn = ConvLearner(None, arch='resnet32') learn.load('stage-1-34') In a production environment where I may want to do single image predictions I don’t want to have to download models.resnet34 since I already have fully trained model weights at my dispos…