README.md

metadata

language:
  - en
license: cc-by-sa-4.0
size_categories:
  - 10K<n<100K
task_categories:
  - image-text-to-text
dataset_info:
  - config_name: MMST_Standard
    description: |
      MIRAGE-MMST Standard Configuration: standard benchmark (train + test).
    citation: |
      @misc{mirage2025,
        title={MIRAGE: A Benchmark for Multimodal Information-Seeking and Reasoning in Agricultural Expert-Guided Conversations},
        author={},
        year={2025},
      }
    features:
      - name: id
        dtype: string
      - name: question
        dtype: string
      - name: answer
        dtype: string
      - name: image_1
        dtype: image
      - name: image_2
        dtype: image
      - name: image_3
        dtype: image
      - name: category
        dtype: string
      - name: entity_type
        dtype: string
      - name: entity_name
        dtype: string
      - name: entity_scientific_name
        dtype: string
      - name: entity_common_names
        list:
          dtype: string
      - name: meta_data_state
        dtype: string
      - name: meta_data_county
        dtype: string
      - name: meta_data_asked_time
        dtype: string
    splits:
      - name: train
        num_examples: 17537
      - name: test
        num_examples: 8188
  - config_name: MMST_Contextual
    description: |
      MIRAGE-MMST Contextual Configuration: contextual benchmark (test only).
    citation: |
      @misc{mirage2025,
        title={MIRAGE: A Benchmark for Multimodal Information-Seeking and Reasoning in Agricultural Expert-Guided Conversations},
        author={},
        year={2025},
      }
    features:
      - name: id
        dtype: string
      - name: question
        dtype: string
      - name: answer
        dtype: string
      - name: image_1
        dtype: image
      - name: image_2
        dtype: image
      - name: image_3
        dtype: image
      - name: category
        dtype: string
      - name: entity_type
        dtype: string
      - name: entity_name
        dtype: string
      - name: entity_scientific_name
        dtype: string
      - name: entity_common_names
        list:
          dtype: string
      - name: meta_data_state
        dtype: string
      - name: meta_data_county
        dtype: string
      - name: meta_data_asked_time
        dtype: string
      - name: location_related
        dtype: bool
      - name: time_related
        dtype: bool
      - name: location_related_analysis
        dtype: string
      - name: time_related_analysis
        dtype: string
    splits:
      - name: test
        num_examples: 3934
  - config_name: MMMT_Direct
    description: >
      MIRAGE-MMMT Direct Configuration: direct-response dialog benchmark with
      three splits (train, dev, test).
    citation: |
      @misc{mirage2025,
        title={MIRAGE: A Benchmark for Multimodal Information-Seeking and Reasoning in Agricultural Expert-Guided Conversations},
        author={},
        year={2025},
      }
    features:
      - name: id
        dtype: string
      - name: dialog_context
        dtype: string
      - name: decision
        dtype: string
      - name: utterance
        dtype: string
      - name: dialog_turns
        dtype: int32
      - name: image_1
        dtype: image
      - name: image_2
        dtype: image
      - name: image_3
        dtype: image
    splits:
      - name: train
        num_examples: 3876
      - name: dev
        num_examples: 878
      - name: test
        num_examples: 861
  - config_name: MMMT_Decomp
    description: >
      MIRAGE-MMMT Decomp Configuration: decomposed-dialog benchmark, with
      known/missing goals.
    citation: |
      @misc{mirage2025,
        title={MIRAGE: A Benchmark for Multimodal Information-Seeking and Reasoning in Agricultural Expert-Guided Conversations},
        author={},
        year={2025},
      }
    features:
      - name: id
        dtype: string
      - name: dialog_context
        dtype: string
      - name: decision
        dtype: string
      - name: utterance
        dtype: string
      - name: dialog_turns
        dtype: int32
      - name: known_goal
        list:
          dtype: string
      - name: missing_goal
        list:
          dtype: string
      - name: image_1
        dtype: image
      - name: image_2
        dtype: image
      - name: image_3
        dtype: image
    splits:
      - name: train
        num_examples: 3876
      - name: dev
        num_examples: 878
      - name: test
        num_examples: 861
configs:
  - config_name: MMST_Standard
    data_files:
      - split: train
        path: MMST_Standard/train/*.arrow
      - split: test
        path: MMST_Standard/test/*.arrow
  - config_name: MMST_Contextual
    data_files:
      - split: test
        path: MMST_Contextual/test/*.arrow
  - config_name: MMMT_Direct
    data_files:
      - split: train
        path: MMMT_Direct/train/*.arrow
      - split: dev
        path: MMMT_Direct/dev/*.arrow
      - split: test
        path: MMMT_Direct/test/*.arrow
  - config_name: MMMT_Decomp
    data_files:
      - split: train
        path: MMMT_Decomp/train/*.arrow
      - split: dev
        path: MMMT_Decomp/dev/*.arrow
      - split: test
        path: MMMT_Decomp/test/*.arrow
modalities:
  - Image
  - Text
tags:
  - biology
  - agriculture
  - Long-Form Question Answering

MIRAGE Benchmark

Project Page | Paper | GitHub

MIRAGE is a benchmark for multimodal expert-level reasoning and decision-making in consultative interaction settings, specifically designed for the agriculture domain. It captures the complexity of expert consultations by combining natural user queries, expert-authored responses, and image-based context.

The benchmark spans diverse crop health, pest diagnosis, and crop management scenarios, including more than 7,000 unique biological entities.

Overview

The benchmark consists of two main components:

• MMST (Multi-Modal Single-Turn): Single-turn multimodal reasoning tasks.
• MMMT (Multi-Modal Multi-Turn): Multi-turn conversational tasks with visual context.

Sample Usage

You can load the various configurations of the dataset using the datasets library:

from datasets import load_dataset

# Load MMST datasets
ds_standard = load_dataset("MIRAGE-Benchmark/MIRAGE", "MMST_Standard")
ds_contextual = load_dataset("MIRAGE-Benchmark/MIRAGE", "MMST_Contextual")

# Load MMMT dataset
ds_mmmt_direct = load_dataset("MIRAGE-Benchmark/MIRAGE", "MMMT_Direct")
ds_mmmt_decomp = load_dataset("MIRAGE-Benchmark/MIRAGE", "MMMT_Decomp")

Citation

If you use our benchmark in your research, please cite our paper:

@article{dongre2025mirage,
  title={MIRAGE: A Benchmark for Multimodal Information-Seeking and Reasoning in Agricultural Expert-Guided Conversations},
  author={Dongre, Vardhan and Gui, Chi and Garg, Shubham and Nayyeri, Hooshang and Tur, Gokhan and Hakkani-T{\"{u}}r, Dilek and Adve, Vikram S},
  journal={arXiv preprint arXiv:2506.20100},
  year={2025}
}

License

This project is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License (CC-BY-SA 4.0).