InfiMM-Eval: Complex Open-ended Reasoning Evaluation for Multi-modal Large Language Models

Logo InfiMM Team 1 2 3
1ByteDance,
2Institute of Automation, Chinese Academy of Sciences, 3University of Texas at Austin
Paper Code Dataset Leaderboard

InfiMM-Eval is an Open-ended VQA benchmark dataset specifically designed for MLLMs, with a focus on complex reasoning tasks.

banner

Overview

Multi-modal Large Language Models (MLLMs) are increasingly prominent in the field of artificial intelligence. These models not only excel in traditional vision-language tasks but also demonstrate im- pressive performance in contemporary multi-modal benchmarks. Although many of these benchmarks attempt to holistically evaluate MLLMs, they typically concentrate on basic reasoning tasks, often yielding only simple yes/no or multi-choice responses. These methods naturally lead to confusion and difficulties in conclusively determining the reasoning capabilities of MLLMs. To mitigate this issue, we manually curate a benchmark dataset specifically designed for MLLMs, with a focus on complex reasoning tasks. Our benchmark comprises three key reasoning categories: deductive, abductive, and analogical reasoning. The queries in our dataset are intentionally constructed to engage the reasoning capabilities of MLLMs in the process of generating answers. For a fair comparison across various MLLMs, we incorporate intermediate reasoning steps into our evaluation criteria. In instances where an MLLM is unable to produce a definitive answer, its reasoning ability is evaluated by requesting intermediate reasoning steps. If these steps align with our manual annotations, appropriate scores are assigned. This evaluation scheme resembles methods commonly used in human assess- ments, such as exams or assignments, and represents what we consider a more effective assessment technique compared with existing benchmarks. We evaluate a selection of representative MLLMs using this rigorously developed open-ended multi-step elaborate reasoning benchmark, designed to challenge and accurately measure their reasoning capabilities.

Statistics

InfiMM-Eval benchmark consists of 279 manually curated reasoning questions, associated with a total of 342 images. The questions are divided into 3 reasoning categories--Deductive, Abductive and Analogical. 49 questions pertain to abductive reasoning, 181 require deductive reasoning, and 49 involve analogicalreasoning. Furthermore, the dataset is divided into two folds based on reasoning complexity, with 108 classified as “High” reasoning complexity and 171 as “Moderate” reasoning complexity.

dataset statistics

Examples

Sample Predictions

Sampled predictions of models on InfiMM-Eval. Model hallucinations are labelled in red.

Leaderboard

MLLM LLM IFT Reasoning Category Reasoning Complexity Overall
Deductive Abductive Analogical Moderate High
OpenFlamingo-v2 MPT-7B No 8.88 5.3 1.11 9.47 4.72 6.82
MiniGPT-v2 LLaMA2-7B Yes 11.02 13.28 5.69 14.45 7.27 10.43
Fuyu-8B Persimmon-8B No 16.42 21.49 7.78 23.06 9.91 15.7
BLIP-2 OPT-2.7B No 22.76 18.96 7.5 24.05 14.18 19.31
InternLM-XComposer-VL InternLM-7B Yes 26.77 35.97 18.61 39.13 17.18 26.84
InstructBLIP FLAN-T5-XXL Yes 27.56 37.76 20.56 40.64 18.09 28.02
LLaMA-Adapter V2 LLaMA-7B No 28.7 46.12 22.08 41.33 21.91 30.46
Otter LLaMA-7B Yes 22.49 33.64 13.33 35.79 12.31 22.69
mPlug-Owl2 LLaMA-7B Yes 23.43 20.6 7.64 28.79 13.18 20.05
IDEFICS-9B-Instruct LLaMA-7B Yes 22.99 34.63 20.56 34.45 16.73 24.53
Emu LLaMA-13B Yes 28.9 36.57 18.19 36.18 22.0 28.24
LLaVA-1.5 Vicuna-13B Yes 30.94 47.91 24.31 47.4 21.0 32.62
CogVLM-Chat Vicuna-7B Yes 36.75 47.88 28.75 55.67 22.5 37.16
Qwen-VL-Chat Qwen-14B Yes 37.55 44.39 30.42 46.61 30.09 37.39
SPHINX v2 LLaMA2-13B Yes 42.17 49.85 20.69 54.85 27.31 39.48
InfiMM-LLaMA13B LLaMA2-13B Yes 41.69 49.70 32.36 61.81 25.09 41.32
GPT-4V GPT-4 Yes 74.86 77.88 69.86 93.98 58.98 74.44

Evaluation

To evaluate on our InfiMM-Eval Benchmark, please follow below steps:

Step 0: Download Images and Questions

Images and Questions can be downloaded HFLogoHere.

Step 1: Generate Response for Your Model

Generate responses for your model on the InfiMM-Eval dataset. The response should be a json file with the following format: 


{ 
  "1": "the answer of question 1", 
  "2": "the answer of question 2", 
  ... 
  "idx": "the answer of question idx" 
}

Step 2: Send Predictions to us

After generating responses for your model, please name the json as model_name_model_size.json e.g. CogVLM-Chat_17B.json and send to us via email for evaluation.

We will evaluate your model and send you the results back.

BibTeX

@misc{han2023coremm,
  title={InfiMM-Eval: Complex Open-Ended Reasoning Evaluation For Multi-Modal Large Language Models}, 
  author={Xiaotian Han and Quanzeng You and Yongfei Liu and Wentao Chen and Huangjie Zheng and Khalil Mrini and Xudong Lin and Yiqi Wang and Bohan Zhai and Jianbo Yuan and Heng Wang and Hongxia Yang},
  year={2023},
  eprint={2311.11567},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}