This post is the second part of my evaluation for the Recurrent Neural Network class. I am going detail what I did during the laboratory.

The lab consisted in implementing a RNN for image captioning, you can find the code on the Google Collab here : https://colab.research.google.com/drive/1qft4rTOrA_FAP6kAPhAMfDisNmcfUh-r?usp=sharing

I completed it with all that we have seen during the class, and I tried to train the model by my own.

As an input of the model , I must give images and their partial caption input, and as the expected output the words to predict.

So first I tried this way :

But I got the following error :

So I added dtype=np.float to the creation of arrays, however I got a new error in my processing data :

It must come from the fact that X is composed of two lists of 2 different sizes

So I converted the partialCaption to the size of the imageInput with padding, but I got a new error while training :

If I do what is asked by the error, I would turn back to the first error. Therefore I did not succeed in training the model with a small dataset.