This project is deprecated and will not be updated anymore. The project continues with this repository: https://github.com/rootvisionai/dml_segmentation

FEWSAM Few-shot Segmentation tool based on Segment Anything

Segment Anything

Installation

SAM requires python>=3.8, as well as pytorch>=1.7 and torchvision>=0.8. Please follow the instructions here to install both PyTorch and TorchVision dependencies. Installing both PyTorch and TorchVision with CUDA support is strongly recommended.

Before you start installation, create an environment first:

conda create --name sam python==3.9

Install Segment Anything:

pip install git+https://github.com/facebookresearch/segment-anything.git

or clone the repository locally and install with

git clone [email protected]:rootvisionai/segment-anything.git
cd segment-anything; pip install -e .

The following dependencies are necessary for the FEWSAM:

pip install opencv-python PyYAML PySimpleGUI

Now download the model checkpoints:

More accurate <<< VIT-H | VIT-L | VIT-B >>> Faster

START SERVER

Before you start the application, create a folder to put your support images that will be used to learn from, then create a folder to put your query images that are going to be labeled. Put the relative path to the folders to support_dir and query_dir in config.yml.

To create request json that will be sent to server

python interface.py  

then adjust make_request.py according to your images and paths

Finally, run the server ...

python backend/server.py

and make request while server.py is running

python make_request.py

DOCKERIZATION

Build docker image

docker build -t few_shot_sam:0.0.1 .

Run docker container

docker run --gpus all -p 80:8080 -it few_shot_sam:0.0.1