Update README.md

README.md

@@ -8,12 +8,14 @@ size_categories:
 - 10K<n<100K
 ---
 # Dataset
-  Two watermark datasets, which can be used for WATERMARK PROTECTION and WATERMARK REMOVAL
+  Two visisble watermark datasets, which can be used for WATERMARK PROTECTION and WATERMARK REMOVAL.
 
 # Dataset Partitioning
   Contains two datasets: LOGO-Multi and LOGO-Full, both of which are newly synthesized based on the LOGO-series datasets. 
+
+  The reference to the LOGO-series datasets is "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."
   
-  The watermarked images in the LOGO-Multi dataset are embedded with multiple watermarks, and the watermarked images in the LOGO-Full dataset are embedded with full watermarks (watermarks cover the entire image).
+  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