Improve dataset card: Add task category, tags, project page, and enhanced description

README.md

@@ -1,9 +1,19 @@
 ---
 license: mit
+task_categories:
+- image-text-to-text
+tags:
+- vision-language
+- spatial-reasoning
+- benchmark
 ---
 
-# Dataset
-For more details, please refer to our [paper](https://arxiv.org/abs/2505.20728).
+# Jigsaw-Puzzles Dataset
+
+Jigsaw-Puzzles is a novel benchmark consisting of 1,100 carefully curated real-world images with high spatial complexity, designed to rigorously evaluate Vision-Language Models' (VLMs) spatial perception, structural understanding, and reasoning capabilities. The dataset minimizes reliance on domain-specific knowledge to better isolate and assess general spatial reasoning, positioning itself as a challenging and diagnostic benchmark for advancing spatial reasoning research in VLMs.
+
+*   **Paper:** [Jigsaw-Puzzles: From Seeing to Understanding to Reasoning in Vision-Language Models](https://arxiv.org/abs/2505.20728)
+*   **Project Page:** https://zesen01.github.io/jigsaw-puzzles
 
 # Citation
 ```