Video Fruit Ripeness Detection App

A small, containerized app that counts fruit of different ripeness on the fly, applying light correction and adjustable a size filter to differentiate background/foreground objects.

Prerequisite: Picking a Fruit Detection Model

Option 1. Using a custom-trained model

Save your model to the path. Change the app/config.yaml file for the correct setup.

# Model configuration
inhouse_model: true
# If using a custom model, set inhouse_model to true and provide the path to the model
model_path: ./app/model.pt

Open source training data can be found the the following Roboflow project.

Option 2. Using a Roboflow pre-trained model

Create a Roboflow account. Get the API for this project: tomato-ripness-unripness under Infer on Local and Hosted Images. Create a ".env" file with the following information:

model_api_key="${YOUR_API_KEY}"

Change the app/config.yaml file for the correct setup.

# Model configuration
inhouse_model: false

Running in local machine

1. Run the shell script in Terminal to set up a virtual environment using Poetry:

source setup.sh

2. Execute the python script:

poetry run python app/app.py

3. Open the user interface at local host: http://127.0.0.1:5000

   Use the buttons below the video stream to:

   • Toggle lighting condition (on: videos corrected to the reference image)
   • Toggle size filter (on: filtering out background fruits as specified in config)
   • Switch between two demo videos

Running in docker environment

1. Build docker image

docker build -t video-fruit-ripeness-detection .

2. Run container

docker run -p 5000:5000 video-fruit-ripeness-detection

3. Open the user interface at local host: http://127.0.0.1:5000

   Use the buttons below the video stream to:

   • Toggle lighting condition (on: videos corrected to the reference image)
   • Toggle size filter (on: filtering out background fruits as specified in config)
   • Switch between two demo videos

Apply the app to custom videos

1. Add your videos to the directory data.

2. Update the experimentation information in app/config.yaml.

3. Run the following to compress your videos before launching the app:

poetry run python app/precompress_videos.py

Project structure

    .
    ├── README.md
    ├── setup.sh          # Shell script to set up a Poetry environment
    ├── poetry.lock      
    ├── pyproject.toml    # Poetry file that defines the python library dependencies
    ├── app              
    │   ├── config*       # Configuration files to specify the default setups
    │   ├── model.pt      # YOLOv8 object detection / instance segmentation model
    │   ├── frontend.html # Frontend to show the user interface
    │   ├── precompress_videos.py        # Script to compress custom videos
    │   ├── app.py        # Flask app to launch the detection service
    │   └── main.py       # Source script to run the inference generator
    └── data              
        ├── *.MP4         # Example video files to make inference from
        └── reference.jpg # reference image for lighting correction