Dog Breed Classification using Machine Learning - Article link : https://medium.com/@judopro/image-recognition-transfer-learning-great-performance-with-limited-data-keras-tensorflow-1299c87f8423
Download data set from http://vision.stanford.edu/aditya86/ImageNetDogs/images.tar
Use ./train/create_validation_set.sh script to create a separate validaton set randomly from within the images and move them to a separate validation folder. Just run it wuthout any parameters, make sure current images are extracted into "train" folder AND you are running this script from withint "train" folder.
Once validation data set is created. You can start the model to train and then predict...
python transfer_learning.py --mode train
python transfer_learning.py --mode predict --image PATH_TO_YOUR_IMAGE.JPG
Your folder structure should be
./doggle_cnn_v8.py
./train/TRAINING_IMAGES <- Run create_validation_set.sh here after download image dataset and expanding into this folder
./validation/VALIDATION_IMAGES