Update dataset card with task overview figure and HF link

README.md

@@ -12,6 +12,8 @@ license_link: LICENSE
 
 VANTAGE-BENCH is the first public benchmark purpose-built for evaluating visual understanding on video captured by fixed infrastructure cameras. It spans three real-world domains — warehouse, smart city / Intelligent Transportation Systems (ITS), and smart spaces — across six spatio-temporal video understanding tasks including video question answering (VQA), temporal grounding, dense video captioning, event verification, spatial grounding, and spatio-temporal tracking.
 
+![VANTAGE-BENCH task overview across Semantic, Temporal, Spatial, and Spatio-Temporal understanding categories](./assets/vantage_bench_tasks.png)
+
 This dataset is for evaluation purposes only.
 
 ## Dataset Owner(s)
@@ -22,7 +24,7 @@ NVIDIA Corporation
 
 April 24, 2026
 
-## License/Terms of Use:
+## License/Terms of Use
 
 [Visit the NVIDIA Legal Release Process](https://nvidia.sharepoint.com/sites/ProductLegalSupport) for instructions on getting legal support for a license selection:
 
@@ -136,7 +138,7 @@ Video (mp4) and Images (jpg).
 
 ## References
 
-**TBD:** Add final HuggingFace repo link.
+- HuggingFace dataset: [nvidia/PhysicalAI-VANTAGE-Bench](https://huggingface.co/datasets/nvidia/PhysicalAI-VANTAGE-Bench)
 
 ## Ethical Considerations