tomhodemon/

grounded-visual-spatial-reasoning

Code for generating the annotations can be found here: github.com

This dataset extends the Visual Spatial Reasoning (VSR) dataset with visual grounding annotations: each caption is annotated with COCO-category object mentions, their positions , and corresponding bounding boxes in the image.

Each sample instance has the following structure:

Field Type Description

image_file string… See the full description on the dataset page: https://huggingface.co/datasets/tomhodemon/grounded-visual-spatial-reasoning.