diff --git a/README.md b/README.md
index 4d1ff3e..b48297d 100644
--- a/README.md
+++ b/README.md
@@ -1,7 +1,16 @@
 # Starbucks
 Starbucks: Improved Training for 2D Matryoshka Embeddings
+<p align="center">
+  <img src="Starbucks.png">
+</p>
 
-### General guidelines
+We propose Starbucks: a new 2D MRL fine-tuning and pre-training method.
+
+Starbucks is composed of two key processes: the Starbucks Masked Autoencoding (SMAE) pretraining and the Starbucks Representation Learning (SRL) fine-tuning processes.
+
+In Starbucks, the  model loss is computed based on a limited target list of layer-dimension pairs, ranging from smaller to larger sizes, much like how the coffeehouses chain [Starbucks](https://en.wikipedia.org/wiki/Starbucks) offers coffee in different cup sizes, from Demi to Trenta.
+
+## General guidelines
 Our codebase is built on top of torch and transformers.
 
 We recommend using a conda environment to install the required dependencies.
@@ -22,7 +31,7 @@ For SRL fine-tuning on retrieval task, see [retrieval](retrieval/README.md).
 
 For SRL fine-tuning on STS task, see [sts](sts/README.md).
 
-### Model Checkpoints
+## Model Checkpoints
 
 We released our model checkpoints on Hugging Face Model Hub: 
 
@@ -30,4 +39,4 @@ Pre-trained SMAE: [bert-base-uncased-fineweb100bt-smae](https://huggingface.co/i
 
 Fine-tuned Starbucks_STS: [Starbucks_STS](https://huggingface.co/ielabgroup/Starbucks_STS)
 
-Fine-tuned Starbucks_Retrieval: [Starbucks_Retrieval](https://huggingface.co/ielabgroup/Starbucks_Retrieval)
\ No newline at end of file
+Fine-tuned Starbucks_Retrieval: [Starbucks-msmarco](https://huggingface.co/ielabgroup/Starbucks-msmarco)
\ No newline at end of file
diff --git a/Starbucks.png b/Starbucks.png
new file mode 100644
index 0000000..7eca473
Binary files /dev/null and b/Starbucks.png differ