NLIP-lab/

LID

This repository hosts the Labelled Illusion Dataset (LID) as a Hugging Face dataset. The images are collected from the AeroCaps and VisDrone datasets. Our train and test sets contain 4,884 and 1,232 samples, respectively. Each sample contains:

Column Name Type Description

image_name string Image filename

image Image Image (embedded in HF datacard)

description string Generated image captions… See the full description on the dataset page: https://huggingface.co/datasets/NLIP-lab/LID.