Improve dataset card: Add description, links, sample usage, and refine metadata

This PR significantly enhances the dataset card for SocialNav-SUB by:
- Adding a descriptive introduction based on the paper's abstract.
- Including direct links to the paper ([https://huggingface.co/papers/2509.08757](https://huggingface.co/papers/2509.08757)), the project page, and the GitHub repository.
- Incorporating a "Getting Started" section with code snippets from the GitHub README for installing dependencies, downloading the dataset, benchmarking a VLM, and viewing results.
- Updating `task_categories` to `question-answering`, `video-text-to-text`, and `robotics`.
- Updating `tags` by removing `robotics` and adding `vqa` for better categorization and searchability.

README.md

@@ -1,13 +1,58 @@
 ---
+language:
+- en
 license: mit
+size_categories:
+- 1K<n<10K
 task_categories:
 - question-answering
-language:
-- en
+- video-text-to-text
+- robotics
 tags:
 - vlm
-- robotics
 - navigation
-size_categories:
-- 1K<n<10K
----
+- vqa
+---
+
+# SocialNav-SUB: Benchmarking VLMs for Scene Understanding in Social Robot Navigation
+
+The Social Navigation Scene Understanding Benchmark (SocialNav-SUB) is a Visual Question Answering (VQA) dataset and benchmark designed to evaluate Vision-Language Models (VLMs) for scene understanding in real-world social robot navigation scenarios. SocialNav-SUB provides a unified framework for evaluating VLMs against human and rule-based baselines across VQA tasks requiring spatial, spatiotemporal, and social reasoning in social robot navigation. It aims to identify critical gaps in the social scene understanding capabilities of current VLMs, setting the stage for further research in foundation models for social robot navigation.
+
+- Paper: [SocialNav-SUB: Benchmarking VLMs for Scene Understanding in Social Robot Navigation](https://huggingface.co/papers/2509.08757)
+- Project Page: [https://larg.github.io/socialnav-sub](https://larg.github.io/socialnav-sub)
+- Code: [https://github.com/michaelmunje/SocialNavSUB](https://github.com/michaelmunje/SocialNavSUB)
+
+## Getting Started
+
+To get started with SocialNav-SUB, you can follow these steps:
+
+1.  **Install Dependencies**
+
+    ```bash
+    pip install -r requirements.txt
+    ```
+
+2.  **Download the Dataset**
+
+    Please download our dataset from [HuggingFace](https://huggingface.co/datasets/michaelmunje/SocialNav-SUB) by running the `download_dataset.sh` script:
+    ```bash
+    ./download_dataset.sh
+    ```
+
+3.  **Benchmark a VLM**
+
+    Make a config file and specify the VLM under the `baseline_model` parameter and parameters for the experiments (such as prompt representation). API models require an environment variable containing an API key (`GOOGLE_API_KEY` or `OPENAI_API_KEY`).
+
+    ```bash
+    python socialnavsub/evaluate_vlm.py --cfg_path <cfg_path>
+    ```
+
+4.  **View Results**
+
+    Results will be saved in the directory specified in the config file under the `evaluation_folder` entry. To postprocess the results, please run:
+
+    ```bash
+    python socialnavsub/postprocess_results.py --cfg_path <cfg_path>
+    ```
+
+    The results will be viewable in the csv whose filepath is specified in the `postprocessed_results_csv` entry in the config file (by default, `postprocessed_results.csv`).