deeplearning4j
How to improve the scores of AnimalsClassification?
I have tried 'Add additional images to the dataset' , 'Increase epochs' and 'Tune by adjusting learning rate, updaters, activation & loss functions' in AnimalsClassification, but the still get very low score. Do you have any suggestions？
It may take a long time to train.
Did you attach a web UI to the running network?
Is it training, is the loss function decreasing?
Did you find hundreds or thousands of images of each class?
The successful image classifiers train for many days on multiple GPU's.
What is the score (accuracy, F1 etc) you are getting? And how much has it improved by the various methods?
Hi everyone,
It would be very interesting to understand for me too how to improve the score of "AnimalClassification.java" because I tried to apply more transforms to dataset, I tried to add more images to the dataset and I tried to increase epochs but I always obtain low scores.
I accept every suggestion.
Thank you.
