config.py

class Config(object):
    def __init__(self):
        ## Questions and Annotataions JSON files
        # self.DATA_DIR ='./datasets/'
        self.DATA_DIR = '/Users/sainikhilmaram/Desktop/datasets/'

        self.TRAIN_QUESTIONS_FILE='v2_OpenEnded_mscoco_train2014_questions.json'
        self.TRAIN_ANNOTATIONS_FILE='v2_mscoco_train2014_annotations.json'
        self.TRAIN_IMAGE_DIR = self.DATA_DIR + '/train2014/'
        self.EVAL_IMAGE_DIR = self.DATA_DIR + '/val2014/'



        self.EVAL_QUESTIONS_FILE='v2_OpenEnded_mscoco_val2014_questions.json'
        self.EVAL_ANNOTATIONS_FILE='v2_mscoco_val2014_annotations.json'

        self.GLOVE_EMBEDDING_FILE='./datasets/glove.6B.100d.txt'

        self.VOCABULARY_FILE = 'vocab_file.csv'

        ## CNN parameters
        self.TRAIN_CNN = False
        self.CNN='vgg16'
        self.CNN_PRETRAINED_FILE = self.DATA_DIR +'./vgg16_weights.npz'
        self.IMAGE_DIMENSION = [224,224]
        self.IMAGE_SHAPE = self.IMAGE_DIMENSION + [3]
        self.IMAGE_FEATURES = 14
        self.IMAGE_FEATURES_MAP = 512
        self.CONV_DATA_SET_TRAIN = 'conv_dict_train.npy'
        self.FC_DATA_SET_TRAIN = 'fc2_dict_train.npy'

        self.CONV_DATA_SET_EVAL = 'conv_dict_eval.npy'
        self.FC_DATA_SET_EVAL = 'fc2_dict_eval.npy'

        # self.CNN = 'resnet50'
        # self.CNN_PRETRAINED_FILE = './resnet50_no_fc.npy'

        ## RNN PARAMETERS
        self.MAX_QUESTION_LENGTH = 25
        self.EMBEDDING_DIMENSION = 512
        self.VOCAB_SIZE = 13764

        ## Decoder Parameters
        self.TOP_ANSWERS = 1000
        self.OUTPUT_SIZE = self.TOP_ANSWERS
        self.TOP_ANSWERS_FILE = 'top_answers.txt'

        self.ONLY_TOP_ANSWERS = True ## If we are considering only questions with top answers in our model
        self.MAX_ANSWER_LENGTH = 1


        ## Model Parameters
        # self.PHASE = 'cnn_features'
        # self.PHASE = 'train'
        self.PHASE = 'test'
        self.EVALUATION_PRESENT = True
        self.POINT_WISE_FEATURES = 1024
        self.INTERMEDIATE_DIMENSION = 30

        self.BATCH_SIZE = 128
        self.INITIAL_LEARNING_RATE = 1e-4
        self.NUM_EPOCHS = 5
        self.NUM_BATCHES = 2 ## Just a place holder, so it doesn't loop through all the data.
        self.SAVE_DIR = './models/'
        self.SAVE_PERIOD = 370000/(self.BATCH_SIZE*4)
        self.LOAD_MODEL = True
        self.MODEL_FILE_NAME= self.SAVE_DIR + '/epoch_14.npy'
        self.EPOCH_COUNT = 5


        ## Testing Parameters
        self.TEST_QUESTION_FILE = 'test_question_file.txt'
        self.TEST_IMAGE_DIR = self.DATA_DIR+'test_image_dir/'


        ## LSTM parameters
        self.LSTM_BATCH_SIZE = self.BATCH_SIZE
        self.LSTM_STEPS = self.MAX_QUESTION_LENGTH
        self.LSTM_CELL_SIZE = self.EMBEDDING_DIMENSION
        self.LSTM_INPUT_SIZE = 32
        self.LSTM_OUTPUT_SIZE = 512
        self.LSTM_DROP_RATE = 0



        # about the weight initialization and regularization
        self.fc_kernel_initializer_scale = 0.08
        self.fc_kernel_regularizer_scale = 1e-4
        self.fc_activity_regularizer_scale = 0.0
        self.conv_kernel_regularizer_scale = 1e-4
        self.conv_activity_regularizer_scale = 0.0
        self.fc_drop_rate = 0.5
        self.lstm_drop_rate = 0.3
        self.attention_loss_factor = 0.01

    def set_batch_size(self,batch_size):
        self.BATCH_SIZE = batch_size
        self.LSTM_BATCH_SIZE = batch_size