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].
To set up the project environment using conda, follow these steps:
git@github.com:aleemsidra/SaLIP.git
cd SaLIP
conda create --name sam python=3.8
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
python main.py --config ./config/lung.json --mode "sam_clip" --prompt_mode "crops" --dataset "cxr" --seed 1234
Feel free to raise an issue or contact me at sidra.aleem2@mail.dcu.ie for queries and discussions.