As part of a deep learning course project, we decide to use neural network to predict film genre from its poster

First if you don't get the posters, run : dataset_preparation.py It will load all images but will take a long time (depending on your internet connection). Then it will create a file "labels.json" which represents a dictionary where keys are name of the films and values are a list of genres of the films.

To create three sets, a training one, a validation one and a test one, run : train_val_test_set_creation.py

You can run python -m unittest test_data_loading to test if all data can be load by Dataset_Manager (in file "dataset_manager.py")

By running finetune.py, the training of the last layer of AlexNet and the finetuning of all its layers are beginning. You now just have to wait, depending on your configuration

When your traning finishes, go to the "logs" folder to see your logs (training loss, mean average precision)

Your model will be saved in saved_models by using early stopping.