Machine-in-the-Loop Rewriting for Creative Image Captioning

Machine-in-the-Loop Rewriting for Creative Image Captioning

Data

Annotated sources of data used in the paper:

TODO: Individual data cleaning scripts

Model Training

Follow the README in the model_training directory to train a Fairseq BART model. Reach out for our trained model.

Interface

Code to run the UI we used for interactive experiments. This UI hosts a server and needs you to have a backend GPU to run model inference during interaction. The code saves each interaction with a unique ID which we use to match to our crowdworkers for experimental analysis.

TODO: Data Processing Scripts to filter results