Fine-tune on Scraped Data

[mattmotoki]

Fine-tune a keras model with a TensorFlow backend (either trained on ImageNet or our PlantNet/Oxford Flower Datasets) using the labels images and labels from data collection.

• create a generator to stream in training data
  • load data in batches
  • perform data augmentation
  • parallelized (optional)
• replace final layer with a fully connected layer with number of neurons equal to the number of plant categories.
• freeze all layers except the last
• train the model with early stopping
• save trained model in TensorflowJS format