Update README.md

README.md

@@ -7,13 +7,41 @@ tags:
 size_categories:
 - 10K<n<100K
 ---
-# Dataset
-  Two new visible watermark datasets, which can be used for WATERMARK PROTECTION and WATERMARK REMOVAL.
+### Summary
 
-# Dataset Partitioning
-  Contains two datasets: LOGO-Multi and LOGO-Full, both of which are newly synthesized based on the LOGO-series datasets ("Cun X, Pun C M. Split then refine: stacked attention-guided resunets for blind single image visible watermark removal[C]//Proceedings of the AAAI conference on artificial intelligence. 2021, 35(2): 1184-1192.")
+This dataset includes two newly synthesized visible watermark datasets intended for:
 
-  The watermarked images in LOGO-Multi contain **multiple** grayscale and colored watermarks, each with different sizes, positions, and transparency; the watermarked images in LOGO-Full contain **a watermark that covers the entire image content**.
-  Each dataset contains 12k samples for training and 2k for testing.
-# Dataset content
-Each sample includes a watermarked image/watermark/mask/original image
+- **Watermark Protection**
+- **Watermark Removal**
+
+---
+
+## Dataset Partitioning
+
+There are **two dataset**:
+
+1. **LOGO-Multi**:  
+   - Contains images with **multiple grayscale and colored watermarks**  
+   - Watermarks vary in **size, position**, and **transparency**
+
+2. **LOGO-Full**:  
+   - Contains images where **a single watermark covers the entire image content**
+   - Watermarks vary in different watermarked images
+
+Each subset includes:
+- **12,000 training samples**
+- **2,000 testing samples**
+
+These two new datasets are constructed based on the LOGO-series datasets described in:  
+*Cun X, Pun C M. “Split then refine: Stacked attention-guided ResUNets for blind single image visible watermark removal.” AAAI Conference on Artificial Intelligence. 2021, 35(2): 1184–1192.*
+
+---
+
+## Dataset Contents
+
+Each sample includes:
+
+- **Watermarked image**
+- **Watermark**
+- **Mask**
+- **Original image**