Add dataset card and link to paper

README.md

@@ -1,3 +1,39 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+task_categories:
+- image-text-to-text
+tags:
+- wildfire
+- vqa
+- thermal-imaging
+- aerial-imagery
+- remote-sensing
+---
+
+# WildFireVQA
+
+WildFireVQA is a large-scale Visual Question Answering (VQA) benchmark for aerial wildfire monitoring that integrates RGB imagery with radiometric thermal data. It is designed to evaluate wildfire-specific multimodal reasoning grounded in temperature measurements for safety-critical scenarios.
+
+- **Paper:** [WildFireVQA: A Large-Scale Radiometric Thermal VQA Benchmark for Aerial Wildfire Monitoring](https://huggingface.co/papers/2604.20190)
+- **Repository:** [https://github.com/mobiiin/WildFire_VQA](https://github.com/mobiiin/WildFire_VQA)
+
+## Dataset Summary
+
+The dataset contains 6,097 RGB-thermal samples. Each sample includes:
+- An RGB image.
+- A color-mapped thermal visualization.
+- A radiometric thermal TIFF file.
+
+These samples are paired with 34 questions each, yielding a total of 207,298 multiple-choice questions. The benchmark evaluates model performance across several operational wildfire intelligence domains:
+- **Presence and detection**
+- **Classification**
+- **Distribution and segmentation**
+- **Localization and direction**
+- **Cross-modal reasoning**
+- **Flight planning**
+
+## Evaluation
+
+The authors provide a unified evaluation pipeline for benchmarking Multimodal Large Language Models (MLLMs) on this dataset. The toolkit supports various open-source VLMs (such as LLaVA, Qwen, and Llama 3.2 Vision) and different input modes (RGB, Thermal, or combined RGB-Thermal).
+
+For detailed instructions on running evaluations, please refer to the [official GitHub repository](https://github.com/mobiiin/WildFire_VQA).