Deep learning for image captioning: Should RNNs be blind?

Corresponding blog post: https://nchlis.github.io/2018_07_28/page.html

The usage of each file is the following:

• load_images.py: reads the Flickr8K images, encodes then with a pre-trained ResNet50 and saves them to disk as Flickr8k_images_encoded.npy. The image filenames are saved into Flickr8k_images_filenames.npy.
• load_captions.py: reads the FLickr8K captions and saves then into captions.npy
• train_model_GRU.py, train_model_LSTM.py: They train a Merge architecture using a GRU and a LSTM, respectively
• train_model_GRU_inject.py, train_model_LSTM.py: same as above, but using Inject architecture
• evaluate_model.py: calculates BLEU scores for a given model
• plot_results.py: plots the results of all models (figures of the blog post comparing the methods)
• generate_new_caption.py: loads a trained model and generates captions for all images present in the ./captioned_images folder