Structural-Damage Image Captioning Dataset

This dataset consists of 2000 images used from (PEER Hub Image-Net) and a caption file in text format.

The caption uses a very limited vocabulary set designed to get better results while also describing plenty of information with a length of 18 vocab.