aleemsidra / SaLIP

This repository contains the pytorch code for our CVPRW 2024 paper "Test-Time Adaptation with SaLIP: A Cascade of SAM and CLIP for Zero-shot Medical Image Segmentation".
MIT License
24 stars 3 forks source link

Test-Time Adaptation with SaLIP: A Cascade of SAM and CLIP for Zero shot Medical Image Segmentation

This project is an implementation of the paper "Test-Time Adaptation with SaLIP: A Cascade of SAM and CLIP for Zero shot Medical Image Segmentation", accepted at [CVPRW 2024].

SaLIP

Setup

To set up the project environment using conda, follow these steps:

  1. Clone the repository: git@github.com:aleemsidra/SaLIP.git
  2. Navigate to the project directory: cd SaLIP
  3. Create a conda environment: conda create --name sam python=3.8
  4. Download SAM's checkpoint from here

Arguments

The main.py script is the main entry point of the project. It performs the following steps:

mode: Specifies the baseline e.g. sam_clip, sam_clip_3d

prompt_mode: Specifies the visual prompt mode for CLIP (default: "crops")

vit_name: Specifies SAM's checkpoint version (default: "vit_h" )

data: Specifies the dataset

config: Specifies the path to config file

seed: Specifies seed value for reproducibility

Example command

python main.py --config ./config/lung.json --mode "sam_clip"  --prompt_mode "crops" --dataset "cxr" --seed 1234 

Contact

Feel free to raise an issue or contact me at sidra.aleem2@mail.dcu.ie for queries and discussions.