Add dataset card, paper link, and GitHub repository for CrashChat

README.md

@@ -0,0 +1,65 @@
+---
+task_categories:
+- video-text-to-text
+language:
+- en
+tags:
+- traffic-safety
+- autonomous-driving
+- video-analysis
+- crash-analysis
+- video-understanding
+---
+
+# CrashChat Dataset
+
+[**Paper**](https://huggingface.co/papers/2512.18878) | [**Code**](https://github.com/Liangkd/CrashChat)
+
+CrashChat is an instruction-tuning crash video dataset specifically designed for multitask traffic crash video analysis. It was developed to train the CrashChat Multimodal Large Language Model (MLLM), enabling automated crash recognition, temporal grounding, and high-level reasoning for traffic safety research.
+
+## Dataset Summary
+
+The dataset encompasses six core tasks related to traffic accident analysis:
+1. **Crash Recognition**: Identifying if a crash occurred.
+2. **Crash Description**: Providing a textual narrative of the event.
+3. **Causal Reasoning**: Explaining why the crash occurred.
+4. **Prevention Reasoning**: Suggesting how the crash could have been avoided.
+5. **Pre-crash Localization**: Identifying the temporal window leading up to the crash.
+6. **Crash Localization**: Grounding the exact timing of the crash event.
+
+### Key Statistics
+- **Total Videos**: 18,385
+- **Total Video-QA Pairs**: 96,184
+
+## Data Structure
+
+According to the official repository, the video data and annotation files are organized as follows:
+
+```bash
+data
+├── videos
+│   ├── cap_1_001537.mp4
+│   ├── cap_1_002004.mp4
+│   └── ...
+├── crashchat_dada_video_total_cause_reasoning_test.json
+├── crashchat_dada_video_total_cause_reasoning_train.json
+├── ...
+└── crashchat_dada_video_total_prevention_reasoning_val.json
+```
+
+## Citation
+
+If you find CrashChat useful for your research, please cite the following paper:
+
+```bibtex
+@article{liang2025crashchat,
+  title   = {CrashChat: A Multimodal Large Language Model for Multitask Traffic Crash Video Analysis},
+  author  = {Liang, Kaidi and Li, Ke and Hu, Xianbiao and Qin, Ruwen},
+  journal = {arXiv preprint arXiv:2512.18878},
+  year    = {2025},
+  archivePrefix = {arXiv},
+  eprint  = {2512.18878},
+  primaryClass = {cs.CV},
+  url     = {https://arxiv.org/abs/2512.18878}
+}
+```