SegmentAnything.jl

Installation

Dependencies

Python

SegmentAnything.jl requires python 3.8.6 with the following packages:

• segment anything
• cv2
• matplotlib.pyplot
• numpy
• torch

You can use pyenv to handle multiple versions of python on your computer. Once pyenv installed, run pyenv install 3.8.6 and pyenv global 3.8.6 to add and to activate the required version on your machine.

There is a requirements.txt file into SegmentAnything.jl/deps/python/. To install the python environment, you just need to create a virtual env and add the packages with the following commands

# change directory 
cd SegmentAnything.jl/deps/python

# create venv
python3 -m venv .venv

# activate venv
source .venv/bin/activate

# install python packages from requirements.txt 
pip3 install -r requirements.txt

SAM

SegmeAnything.jl needs the Segment Anything Model Checkpoints. To download the 3 checkpoints, just run the download_sam.jl file located at SegmentAnything/deps/sam.

SegmentAnything.jl

• Clone the repo
• In a terminal emulator, change the working directory: cd path/to/SegmentAnything.jl
• Activate the julia env in the julia REPL: pkg> activate.
• Instantiate the env to install deps: pkg> instantiate
• Then, load the module: julia> include("SegmentAnything.jl")

Getting Started

Automatic segmentation

To perform an automatic segmentation, just run the segment_image() function. This function takes two arguments, the image path and a sam checkpoint (vit_b, vit_l or vit_h).

julia> image = "path/to/image"
julia> segmentation =  SegmentAnything.segment_image(image, "vit_b")

To display the masks over image, use the show_anns_jl() function.

julia> SegmentAnything.show_anns_jl(segmentation, image)