<aside> 💡 In den Notebooks 4-5 werden best practices angewendet und viele Konzepte erklärt.
</aside>
Preprocessing
Fastai preprocessing example for images. Use fastcore.parallel to make it faster. Afterwards transform images. Use show_batch to check your images.
Choose model
To pick an architecture, we should look at the options in The best vision models for fine-tuning (see fastai tutorial). In the linked table you can find the best model to start fine-tuning. At the beginning we take the one which is the fastes to train. Here is another statement about this table.
Learning Rate finder
After picking a model, start to learn with vision_learner and see what the learning rate finder lr_find shows.
Predictions
Get first predictions. We want to do exactly the same transformations to the test set.
Try out different architecture