Low accuracy of LRCN model

Hello,

I'm trying to recreate the results you've mentioned in your medium blog post. I've split the dataset using your scripts, used the default hyper parameters- 1000 epochs lr = 1e-5, decay - 1e-6 Adam optimizer

But after 60 epochs or so, the results are as follows - (very low accuracy) 131/131 [==============================] - 1315s 10s/step - loss: 3.8444 - acc: 0.1359 - top_k_categorical_accuracy: 0.3930 - val_loss: 4.3090 - val_acc: 0.0922 - val_top_k_categorical_accuracy: 0.2879

Can you please clarify if I'm missing something, or I'm supposed to run for 500+ epochs? thanks