by Ma Lab
MedSAM is a repository for segmenting medical images, adapted from Segment Anything. You can run it from the command line, a notebook or a desktop GUI.
Web appFree
Step 1: Create a conda environment with Python 3.10 and install PyTorch 2.0.
Step 2: Clone the repository and run pip install -e . inside it.
Step 3: Download the model checkpoint and place it in the work_dir folder.
Step 4: Run MedSAM_Inference.py with input, output and bounding box flags.
Step 5: Or open the notebook tutorial, or start gui.py and draw boxes.
You specify the target with a bounding box, either by flag on the command line or by drawing it in the GUI.
Yes. The README gives preprocessing steps and training scripts for one GPU or multiple GPUs.
The README states five A100 nodes with four 80G GPUs each, 20 A100 GPUs in total.
The news lists LiteMedSAM with a 3D Slicer plugin, and MedSAM2 for 3D and video segmentation.
nnU-NetnnU-Net is an open-source segmentation framework that analyzes a dataset and configures its own U-Net pipeline, from preprocessing to training and inference. It is mainly used for biomedical image segmentation.
PubMedCLIPMedical vision-language model adapting CLIP to radiology and biomedical image-text understanding
MedCLIPMedCLIP is a Python package for contrastive learning from unpaired medical images and texts, published at EMNLP 2022. It provides pretrained vision-text models loadable in a few lines.