Improve model card: Add library_name, GitHub link, and usage example

README.md

@@ -2,12 +2,17 @@
 pipeline_tag: image-segmentation
 tags:
 - medical
+license: mit
+library_name: transformers
 ---
+
 # MCP-MedSAM
 
 Pytorch Implementation of the paper:
 "[MCP-MedSAM: A Powerful Lightweight Medical Segment Anything Model Trained with a Single GPU in Just One Day](https://arxiv.org/abs/2412.05888)"
 
+Code: [https://github.com/Leo-Lyu/MCP-MedSAM](https://github.com/Leo-Lyu/MCP-MedSAM)
+
 ![MCP-MedSAM Architecture](docs/MCP-MedSAM.png)
 
 ## 📄 Overview
@@ -24,11 +29,6 @@ To further improve performance across imaging modalities, we introduce a **modal
 
 With these enhancements, our model achieves strong multi-modality segmentation performance, and can be trained in approximately **1 day on a single A100 (40GB)** GPU.
 
-<!-- 
-We are currently releasing the inference code along with the model weight. You can download from [here](https://drive.google.com/drive/folders/1NW4aSNhk-dtiK-dicTAUp0g0eR2fryNi?usp=sharing).
-
-The training code has been released and you can train your . -->
-
 ## Requirements
 
 * Python==3.10.14
@@ -38,7 +38,42 @@ The training code has been released and you can train your . -->
 
 ## Training and Inference
 
-Training and inference can be done by running train.py and infer.py. Model weights are stored in the pytorch_model.bin file, which can be loaded for inference.
+Training and inference can be done by running `train.py` and `infer.py` from the [official repository](https://github.com/Leo-Lyu/MCP-MedSAM). Additionally, the model weight for inference can be downloaded from [here](https://drive.google.com/drive/folders/1NW4aSNhk-dtiK-dicTAUp0g0eR2fryNi?usp=sharing). MCP-MedSAM has also been uploaded to the [Hugging Face Hub](https://huggingface.co/Leo-Lyu/MCP-MedSAM), including pre-trained weights as well.
+
+## Usage
+
+You can use the model with the `transformers` library for inference. Ensure you have the `transformers` library installed (`pip install transformers`).
+
+```python
+from transformers import AutoModelForImageSegmentation, AutoProcessor
+from PIL import Image
+import torch
+
+# Load model and processor from the Hugging Face Hub
+model = AutoModelForImageSegmentation.from_pretrained("Leo-Lyu/MCP-MedSAM")
+processor = AutoProcessor.from_pretrained("Leo-Lyu/MCP-MedSAM")
+
+# Example: Load an image and define a bounding box prompt
+# Replace "path/to/your/medical_image.jpg" with the actual path to your image file
+image = Image.open("path/to/your/medical_image.jpg").convert("RGB")
+input_boxes = [[100, 200, 300, 400]] # Example bounding box prompt [x_min, y_min, x_max, y_max]
+
+# Prepare inputs for the model
+inputs = processor(images=image, input_boxes=input_boxes, return_tensors="pt")
+
+# Perform inference
+with torch.no_grad():
+    outputs = model(**inputs)
+
+# Access predicted masks (raw logits)
+# The exact output structure might vary depending on the model implementation.
+pred_masks = outputs.pred_masks
+
+# To get a binary mask, apply sigmoid and threshold (example)
+binary_mask = (torch.sigmoid(pred_masks) > 0.5).float()
+
+print("Generated mask shape:", binary_mask.shape)
+```
 
 ## Citation