Update README.md

README.md

@@ -24,18 +24,6 @@ The dataset contains:
 - **Two modalities**: `image` or `video`
 - **Two question formats**: `multiple-choice` (MC) or `true/false` (TF)
 
----
-
-## 🧩 Reasoning Types
-
-The benchmark questions challenge perceptual consistency through:
-
-- Object identity under occlusion
-- Shape/color constancy
-- Spatial invariance
-- Visual illusions
-- Comparative reasoning across altered perspectives
-- Common physical scene understanding
 
 ---
 
@@ -77,7 +65,7 @@ data/
 ```
 
 - The `metadata.jsonl` files store structured entries for each modality.
-- `.gif` files are stored in `videos/` and marked as `media_type = video` to ensure compatibility with Dataset Viewer.
+- `.gif` files are stored in `videos/` and marked as `media_type = video`.
 
 ---
 
@@ -85,12 +73,12 @@ data/
 
 ```json
 {
-  "file_name": "a0044.gif",
-  "media_type": "video",
+  "file_name": "a0033.JPG",
+  "media_type": "image",
   "question_type": "TF",
-  "sec_label": 3,
-  "question": "Are the two monsters in the picture actually the same size?",
-  "correct_answer": "Yes"
+  "sec_label": 1.0,
+  "question": "In the picture, has the actual color of the bridge itself changed?",
+  "correct_answer": "No"
 }
 ```
 
@@ -101,11 +89,14 @@ data/
 If you use this dataset, please cite:
 
 ```bibtex
-@misc{sun2025perceptualconstancy,
-      title={Probing Perceptual Constancy in Vision-Language Models}, 
-      author={Haoran Sun and Qingying Gao and Haiyun Lyu and Dezhi Luo and Yijiang Li and Hokin Deng},
+@misc{sun2025probingperceptualconstancylarge,
+      title={Probing Perceptual Constancy in Large Vision Language Models}, 
+      author={Haoran Sun and Suyang Yu and Yijiang Li and Qingying Gao and Haiyun Lyu and Hokin Deng and Dezhi Luo},
       year={2025},
-      note={In preparation}
+      eprint={2502.10273},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2502.10273}, 
 }
 ```
 
@@ -113,4 +104,4 @@ If you use this dataset, please cite:
 
 ## 🤝 Acknowledgments
 
-This dataset is developed by the Grow AI Like a Child community to support structured, curriculum-aligned evaluation of visual invariance and high-level perceptual reasoning.
+This dataset is developed by the Grow AI Like a Child community to support structured.