Datasets:

etri-lirs
/

SFE

@@ -28,68 +28,22 @@ configs:
     path: "test.zh.parquet"
 ---
-# <center>Scientists' First Exam: Probing Cognitive Abilities of MLLM via Perception, Understanding, and Reasoning</center>
-<p align="center">
-  <img src="./assets/sfe_title.png" alt="SFE" style="display: block; margin: auto; max-width: 100%;">
-</p>
-<p align="center">
-| <a href="https://prismax.opencompass.org.cn/vlmlb"><b>Leaderboard</b></a> | <a href="https://arxiv.org/abs/2506.10521"><b>Paper</b> </a> | <a href="https://prismax.opencompass.org.cn/"><b>Website</b> </a> | <a href="https://huggingface.co/datasets/PrismaX/SFE"><b>HuggingFace</b> </a> |
-</p>
----
-*Latest News* 🔥
-[Latest] [Seed-1.8](https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/research/Seed-1.8-Modelcard.pdf), the model with native support for generalized real-world agency, is benchmarked on SFE.
-<details>
-<summary>Unfold to see more details.</summary>
-<be>
-- [2025/12] [Seed-1.8](https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/research/Seed-1.8-Modelcard.pdf), the model with native support for generalized real-world agency, is benchmarked on SFE.
-- [2025/07] [Intern-S1](https://github.com/InternLM/Intern-S1), the most advanced open-source multimodal reasoning model to date, is benchmarked on SFE.
-- [2025/07] We are officially integrated by [VLMEvalKit](https://github.com/open-compass/VLMEvalKit).
-- [2025/06] We officially released SFE! SFE is designed to evaluate the scientific cognitive capacities of MLLMs through three cognitive levels: **scientific signal perception**, **scientific attribute understanding**, and **scientific comparative reasoning**.
-</details>
----
-## Motivation: Current scientific benchmarks inadequately assess MLLMs
-<details>
-<summary>Unfold to see more details.</summary>
-<br>
-Scientific discoveries increasingly rely on complex multimodal reasoning based on information-intensive scientific data and domain-specific expertise. Empowered by expert-level scientific benchmarks, scientific Multimodal Large Language Models (MLLMs) hold the potential to significantly enhance this discovery process in realistic workflows. However, current scientific benchmarks mostly focus on evaluating the knowledge understanding capabilities of MLLMs, leading to an inadequate assessment of their perception and reasoning abilities. To address this gap, we present the Scientists’ First Exam (SFE) benchmark, designed to evaluate the scientific cognitive capacities of MLLMs through three interconnected levels: **scientific signal perception**, **scientific attribute understanding**, **scientific comparative reasoning**. Specifically, SFE comprises 830 expert-verified VQA pairs across three question types, spanning 66 multimodal tasks across five high-value disciplines. Extensive experiments reveal that current **state-of-the-art** GPT-o3 and InternVL-3 achieve only 34.08% and 26.52% on SFE, highlighting significant room for MLLMs to improve in scientific realms. We hope the insights obtained in SFE will facilitate further developments in AI-enhanced scientific discoveries.
-</details>
-## Overview
-<p align="center">
-  <img src="./assets/sfe_distribution.png" alt="The structure of SFE includes 5 disciplines, 18 scientific directions, and 66 tasks." style="display: block; margin: auto; max-width: 50%;">
-</p>
-We introduce the Scientists' First Exam (SFE) benchmark, designed to comprehensively evaluate the scientific cognitive capabilities of MLLMs through three cognitive levels (cog-levels):
-1. **Scientific Signal Perception** characterizes the capacity to discern critical components within visualizations of scientific raw data.
-2. **Scientific Attribute Understanding** demonstrates the ability to interpret domain-expert knowledge.
-3. **Scientific Comparative Reasoning** manifests the ability to derive phenomenological insights through structured comparison of multiple scientific visual sources.
-SFE encompasses 66 expert-curated, high-value multimodal tasks across five disciplines: Astronomy, Chemistry, Earth, Life, and Materials Sciences.
-Each task is constructed from native scientific raw data formats and formulated as visual question answering (VQA) pairs, designed to probe specific levels of scientific cognition.
-All tasks are bilingual (English \& Chinese) to support broad accessibility.
-These tasks are designed not only to require a deep understanding of domain-specific knowledge and data analysis skills but also to significantly enhance research efficiency and facilitate advancements that benefit society.
-## Download Dataset
-```bash
-git lfs install
-git clone https://huggingface.co/datasets/PrismaX/SFE # Clone all files, including raw data
-GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/PrismaX/SFE # If you want to clone without large files - just their pointers
-```
 ## Evaluations

     path: "test.zh.parquet"
 ---
+## Repackaging Notice / 재패키징 안내 / 重新打包说明
+**English**
+This dataset is a repackaged version of the multimodal dataset [`InternScience/SFE`](https://huggingface.co/datasets/InternScience/SFE), based on its 2025-06-11 release. The original content has been preserved, while all images have been embedded directly into the dataset files to simplify usage and distribution on the Hugging Face Hub.
+For the original dataset description, license, and citation, please refer to the source dataset.
+**한국어 (Korean)**
+이 데이터셋은 멀티모달 데이터셋 [`InternScience/SFE`](https://huggingface.co/datasets/InternScience/SFE)의 2025-06-11 버전을 기반으로 재패키징한 것입니다. 원본 데이터 내용은 유지하면서, Hugging Face Hub에서의 사용 편의성을 위해 모든 이미지를 데이터셋 내부에 임베딩했습니다.
+데이터셋 설명, 라이선스 및 인용 정보는 원본 데이터셋을 참고하시기 바랍니다.
+**中文 (Chinese)**
+本数据集基于多模态数据集 [`InternScience/SFE`](https://huggingface.co/datasets/InternScience/SFE) 的 2025-06-11 版本进行重新打包。在保持原始数据内容不变的前提下，我们将所有图像直接嵌入到数据文件中，以便在 Hugging Face Hub 上更方便地使用和分发。
+有关原始数据集的描述、许可证和引用信息，请参考源数据集。
+# <center>Scientists' First Exam: Probing Cognitive Abilities of MLLM via Perception, Understanding, and Reasoning</center>
 ## Evaluations