Add comprehensive dataset card for MISS-QA

README.md

@@ -0,0 +1,90 @@
+---
+task_categories:
+- image-text-to-text
+language:
+- en
+tags:
+- question-answering
+- multimodal
+- scientific-literature
+- schematic-diagrams
+- benchmark
+---
+
+# MISS-QA: A Multimodal Scientific Information-Seeking QA Benchmark
+
+MISS-QA (Multimodal Information-Seeking over Scientific papers – Question Answering) is the first benchmark specifically designed to evaluate the ability of multimodal foundation models to interpret schematic diagrams and answer information-seeking questions within scientific literature.
+
+This dataset is associated with the paper: [Can Multimodal Foundation Models Understand Schematic Diagrams? An Empirical Study on Information-Seeking QA over Scientific Papers](https://huggingface.co/papers/2507.10787).
+
+Code: [https://github.com/QDRhhhh/MISSQA](https://github.com/QDRhhhh/MISSQA)
+
+## Highlights
+
+- **1500 QA pairs** annotated by **expert researchers**.
+- Covers **465 AI-related papers** from arXiv.
+- Focuses on **schematic diagrams**, not just charts or tables.
+- Evaluates **18 frontier vision-language models** (o4-mini, Gemini-2.5-Flash, and Qwen2.5-VL).
+- Automatic evaluation protocol trained on **human-scored data**.
+
+## Benchmark Structure
+
+Each example in MISS-QA includes:
+
+- A **schematic diagram** from a scientific paper.
+- A **highlighted visual element** (bounding box).
+- A **free-form information-seeking question**.
+- The corresponding **scientific context**.
+- A human-annotated **answer** (or marked as unanswerable).
+
+### Information-Seeking Scenarios
+
+- **Design Rationale**
+- **Implementation Details**
+- **Literature Background**
+- **Experimental Results**
+- **Other** (e.g., limitations, ethics)
+
+## Model Evaluation
+
+MISS-QA is used to benchmark proprietary and open-source **multimodal foundation models**. Performance is automatically scored using a custom evaluation protocol aligned with human judgment.
+
+## How to Use
+
+### Step 0: Installation
+
+```bash
+git clone https://github.com/QDRhhhh/MISSQA.git
+cd MISSQA
+conda create --name missqa python=3.10
+conda activate missqa
+pip install -r requirements.txt
+```
+
+### Step 1: Run Model Inference
+
+Use the provided bash script to run inference with your multimodal model:
+
+```bash
+bash scripts/vllm_large.sh
+```
+
+This will generate model responses and save them to:
+
+```swift
+./outputs/
+```
+
+### Step 2: Evaluate Model Accuracy
+
+Once inference is complete, run the accuracy evaluation script:
+
+```bash
+python acc_evaluation.py
+```
+
+The processed and scored outputs will be saved to:
+
+```swift
+./processed_outputs/
+```