Update README.md

README.md

@@ -27,12 +27,12 @@ Our research reveals that issue lies not only with the models but with the bench
 On this enhanced benchmark, state-of-the-art models achieve success rates nearing 80% on complex tasks, reflecting that on-device GUI agents are actually closer to practical deployment than previously thought. We also trained our new SOTA model, **Magma-R1**, on just 2,400 curated samples, which matches the performance of previous models trained on over 31,000 samples.
 
 <div align="center">
-  <img src="static/images/method_1021_1355-compress.png" width="90%" alt="Method Overview">
+  <img src="static/images/method_1013_1355-compress.png" width="90%" alt="Method Overview">
   <p><i>Overview of our integrated pipeline for Magma-R1 training and AndroidControl-Curated creation.</i></p>
 </div>
 
 ## 🔥 News
-- 🔥 ***`2025/10/21`*** Our paper "[AndroidControl-Curated: Revealing the True Potential of GUI Agents through Benchmark Purification](YOUR_ARXIV_PAPER_LINK)" released.
+- 🔥 ***`2025/10/21`*** Our paper "[AndroidControl-Curated: Revealing the True Potential of GUI Agents through Benchmark Purification](https://arxiv.org/abs/2510.18488)" released.
 
 ## 🚀 Updates
 - ***`2025/10/21`*** The source code for `AndroidControl-Curated` and `Magma-R1` has been released.
@@ -74,8 +74,8 @@ On this enhanced benchmark, state-of-the-art models achieve success rates nearin
 
 1.  **Clone the repository:**
     ```bash
-    git clone https://github.com/YourUsername/YourRepoName.git
-    cd YourRepoName
+    git clone https://github.com/batechworks/AndroidControl_Curated.git
+    cd AndroidControl_Curated
     ```
 
 2.  **Install dependencies:**
@@ -89,7 +89,7 @@ On this enhanced benchmark, state-of-the-art models achieve success rates nearin
 To reproduce the results on `AndroidControl-Curated`:
 
 1.  **Download the benchmark data:**
-    Download the processed test set from [Hugging Face](YOUR_HUGGINGFACE_DATASET_LINK) and place it in the `benchmark_resource/` directory. The directory should contain the following files:
+    Download the processed test set from [Hugging Face](https://huggingface.co/datasets/batwBMW/AndroidControl_Curated) and place it in the `benchmark_resource/` directory. The directory should contain the following files:
     - `android_control_high_bbox.json`
     - `android_control_high_point.json`
     - `android_control_low_bbox.json`
@@ -97,7 +97,7 @@ To reproduce the results on `AndroidControl-Curated`:
     - `android_control_high_task-improved.json`
 
 2.  **Download the model:**
-    Download the `Magma-R1` model weights from [Hugging Face](YOUR_HUGGINGFACE_MODEL_LINK) and place them in your desired location.
+    Download the `Magma-R1` model weights from [Hugging Face](https://huggingface.co/batwBMW/Magma-R1) and place them in your desired location.
 
 3.  **Run the evaluation script:**
     Execute the following command, making sure to update the paths to your model and the benchmark image directory.