This benchmark is used to train and evaluate the Continual Instruction Tuning capabilities of MLLMs and is proposed by HiDe-LLaVA (ACL 2025).
This repository contains mainly the training and testing instructions for the datasets used as well as images of ImageNet-R and Flickr30k datasets. For images of other datasets, please refer to the links provided in our GitHub.
If you use our benchmarks, please cite our work：
@article{guo2025hide,
title={Hide-llava:… See the full description on the dataset page:
https://huggingface.co/datasets/HaiyangGuo/UCIT.
