# screen2words Dataset

The dataset contains 112,085 screen summaries in CSV format for 22,417 screens from the RICO dataset. The split/ folder contains the train/dev/test splits used by the paper for model training and evaluation.