From the course: Transfer Learning for Images Using PyTorch: Essential Training

Unlock the full course today

Join today to access over 22,600 courses taught by industry experts or purchase this course individually.

Next steps

Next steps

From the course: Transfer Learning for Images Using PyTorch: Essential Training

Start my 1-month free trial

Next steps

“

- [Jonathan] So, if you're new to Transfer Learning, then you've learned a couple of new techniques. Go ahead and try Transfer Learning on your own dataset. Now in training, go ahead and try three, five, 10, or maybe even 20 epochs over the training data. For the sake of time, I've mostly demonstrated training over one epoch. But you can see that we got an accuracy of around 80% off the bat. Now I'd suggest getting familiar with some of the trained deep learning models. If I have to do a quick computer-vision prototype, my go-to model for a balance of speed and accuracy is RasNet50. I'd normally try a fixed-feature extractor to get a baseline accuracy, and then I'd try fine tuning to see what results I get wth three to five epochs. If I can get an additional two to 3% accuracy with learning rates or differential learning rates, I'd normally train the model anywhere between five and 20 epochs. You might find that…

Contents