Collaborative Instance Object Navigation - CoIN-Bench dataset (v1) - ICCV 25

We introduce Collaborative Instance object Navigation (CoIN), a new task setting where the agent actively resolve uncertainties about the target instance during navigation in natural, template-free, open-ended dialogues with human.

To download the dataset, just run the following command:

from huggingface_hub import snapshot_download
snapshot_download(repo_id="ftaioli/CoIN-Bench", repo_type="dataset", local_dir="CoIN-Bench")

Dataset Details

Please see our ICCV 25 accepted paper: Collaborative Instance Object Navigation: Leveraging Uncertainty-Awareness to Minimize Human-Agent Dialogues.

For more information, visit our Github repo.

Dataset Structure

This repository contains the CoIN-Bench dataset, which is structured as follows:

• val_unseen: Contains only novel objects not present in the training set.
• val_seen: Includes objects that also appear in the training set.
• val_seen_synonyms: Contains objects from the training set but with synonymous names.

Dataset Description

• Curated by: Francesco Taioli and Edoardo Zorzi.

Citation

@misc{taioli2025collaborativeinstanceobjectnavigation,
      title={Collaborative Instance Object Navigation: Leveraging Uncertainty-Awareness to Minimize Human-Agent Dialogues}, 
      author={Francesco Taioli and Edoardo Zorzi and Gianni Franchi and Alberto Castellini and Alessandro Farinelli and Marco Cristani and Yiming Wang},
      year={2025},
      eprint={2412.01250},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2412.01250}, 
}