Fast use

docker pull pablogod/fishcut
docker container run -p 8501:8501 --rm -d pablogod/fishcut

Fish YOLO GrabCut

YOLOv3 object detection then use GrabCut to do semantic segmentation to fish market images.

Demo on Heroku

I have deployed this project on Heroku. You can try here: https://fish-yolo-grabcut.herokuapp.com/

Set up and Run Demo Locally

First, create a virtual environment and install the dependencies (assume you are using conda):

$ conda create -n fish-yolo-grabcut python=3.6 pip 
$ conda activate fish-yolo-grabcut
$ pip install -r requirements.txt

Then, use gdown to download the pretrained weights from here and put it to yolo-fish directory:

$ gdown --output ./yolo-fish/fish.weights --id 1M8dKL0mjh5QkdH2UeFQN9RF3pXCV6hao

Next, choose one of the following approaches you like.

1. Command Line Approach

$ python main.py --image ./images/DSC_0061.JPG --yolo yolo-fish

When finishing, you should find 8 jpg images in the project root directory.

2. Streamlit Approach

$ streamlit run app.py

You can upload fish market image to run the program.

The results are shown in the browser (make sure to scroll down).

3. Docker Approach

$ docker image build -t fish-yolo-grabcut:app .
$ docker container run -p 8501:8501 --rm -d grabcut:app

Then upload fish market image to run the program.

The results are shown in the browser (make sure to scroll down).

To shutdown the docker type this:

$ docker kill <weird id of fish-yolo-grabcut.app>