Image Captioning Approaches

Build and experiment scalable Image Captioning techniques.

Dataset Loaders

Data preparation and loading is complete for the following Image captioning datasets

1. Flickr8k
2. Google Conceptual Captions

Download Conceptual Captions

1. The main dataset is available for download at https://ai.google.com/research/ConceptualCaptions/download

Steps to download images:

1. Download the training split from the dataset page
2. Set the GOOGLE_CAPTIONS_FILE url path in preprocessing\GoogleCaptions.py.
3. Choose the number of samples to download by setting NUM_SAMPLES
4. Run the script using python3 GoocleCaptions.py
5. The images would be downloaded in the set directory

Keras Loaders

1. The keras data loaders are written for Flickr8k and Google Conceptual Captions at datasets\flickr8k and datasets\googlecc.py.
2. Currently, there is no separate usage of Validation set and a part of the training data is used for Validation.

Training