Update README.md

README.md

@@ -1,2 +1,43 @@
-Source: Serre Lab at Brown University
-https://serre-lab.clps.brown.edu/
+---
+annotations_creators:
+  - expert-generated
+language:
+  - und
+multilingual: false
+pretty_name: HMDB51
+task_categories:
+  - video-classification
+task_ids:
+  - action-recognition
+license: other  # HMDB51 没有明确开源协议，可以注明为 "other"
+dataset_creators:
+  - Serre Lab at Brown University
+paperswithcode_id: hmdb51
+tags:
+  - human-action
+  - video
+  - benchmark
+---
+
+# HMDB51 Dataset
+
+**HMDB51** is a large human action recognition dataset consisting of 6,766 video clips distributed across 51 action categories (e.g., jumping, drinking, kicking, etc). It is commonly used as a benchmark for video classification and action recognition tasks.
+
+## 📦 Dataset Source
+
+This dataset was originally collected and published by the **Serre Lab at Brown University**.
+
+🔗 [Serre Lab – Brown University](https://serre-lab.clps.brown.edu/)
+
+## 📁 Contents
+
+- 6,766 video clips
+- 51 action categories
+- Includes training/testing splits for benchmarking
+
+## 📚 Citation
+
+If you use this dataset, please cite:
+
+> Kuehne, H., Jhuang, H., Garrote, E., Poggio, T., & Serre, T. (2011). HMDB: A large video database for human motion recognition. In Proceedings of the IEEE International Conference on Computer Vision (ICCV).
+