training model description

training_model.py

"""
time_steps_encoder is the number of frames per video we will be using for training
num_encoder_tokens is the number of features from each frame
latent_dim is the number of hidden features for lstm
time_steps_decoder is the maximum length of each sentence
num_decoder_tokens is the final number of tokens in the softmax layer
batch size
"""
time_steps_encoder=80
num_encoder_tokens=4096
latent_dim=512
time_steps_decoder=10
num_decoder_tokens=1500
batch_size=320

encoder_inputs = Input(shape=(time_steps_encoder, num_encoder_tokens), name="encoder_inputs")
encoder = LSTM(latent_dim, return_state=True,return_sequences=True, name='endcoder_lstm')
_, state_h, state_c = encoder(encoder_inputs)
encoder_states = [state_h, state_c]


# Set up the decoder
decoder_inputs = Input(shape=(time_steps_decoder, num_decoder_tokens), name= "decoder_inputs")
decoder_lstm = LSTM(latent_dim, return_sequences=True, return_state=True, name='decoder_lstm')
decoder_outputs, _, _ = decoder_lstm(decoder_inputs, initial_state=encoder_states)
decoder_dense = Dense(num_decoder_tokens, activation='softmax', name='decoder_relu')
decoder_outputs = decoder_dense(decoder_outputs)


model = Model([encoder_inputs, decoder_inputs], decoder_outputs)
model.summary()