README.md

metadata

language:
  - en
license: apache-2.0
size_categories:
  - n<1K
task_categories:
  - question-answering
pretty_name: Mantis-Eval
dataset_info:
  - config_name: mantis_eval
    features:
      - name: id
        dtype: string
      - name: question_type
        dtype: string
      - name: question
        dtype: string
      - name: images
        sequence: image
      - name: options
        sequence: string
      - name: answer
        dtype: string
      - name: data_source
        dtype: string
      - name: category
        dtype: string
    splits:
      - name: test
        num_bytes: 479770102
        num_examples: 217
    download_size: 473031413
    dataset_size: 479770102
configs:
  - config_name: mantis_eval
    data_files:
      - split: test
        path: mantis_eval/test-*

Overview

This is a newly curated dataset to evaluate multimodal language models' capability to reason over multiple images. More details are shown in https://tiger-ai-lab.github.io/Mantis/.

Statistics

This evaluation dataset contains 217 human-annotated challenging multi-image reasoning problems.

Leaderboard

We list the current results as follows:

Models	Size	Mantis-Eval
LLaVA OneVision	72B	77.60
LLaVA OneVision	7B	64.20
GPT-4V	-	62.67
Mantis-SigLIP	8B	59.45
Mantis-Idefics2	8B	57.14
Mantis-CLIP	8B	55.76
VILA	8B	51.15
BLIP-2	13B	49.77
Idefics2	8B	48.85
InstructBLIP	13B	45.62
LLaVA-V1.6	7B	45.62
CogVLM	17B	45.16
LLaVA OneVision	0.5B	39.60
Qwen-VL-Chat	7B	39.17
Emu2-Chat	37B	37.79
VideoLLaVA	7B	35.04
Mantis-Flamingo	9B	32.72
LLaVA-v1.5	7B	31.34
Kosmos2	1.6B	30.41
Idefics1	9B	28.11
Fuyu	8B	27.19
OpenFlamingo	9B	12.44
Otter-Image	9B	14.29

Citation

If you are using this dataset, please cite our work with

@inproceedings{Jiang2024MANTISIM,
  title={MANTIS: Interleaved Multi-Image Instruction Tuning},
  author={Dongfu Jiang and Xuan He and Huaye Zeng and Cong Wei and Max W.F. Ku and Qian Liu and Wenhu Chen},
  publisher={arXiv2405.01483}
  year={2024},
}