BirdFSD-YOLOv5

Build and train a custom model to identify birds visiting bird feeders.

📖 Documentation

Requirements

• 🐍 python>=3.8

🚀 Getting started

• First, fork the repository.
• Enable workflows in your fork:

• Then, click on and enable all the workflows that are highlighted wuth a red square in the image below.

• On your machine, run:

git clone https://github.com/bird-feeder/BirdFSD-YOLOv5.git
cd BirdFSD-YOLOv5

# Recommended: create a conda/pyenv environment
pip install -r requirements.txt

poetry build
pip install dist/*.whl

git clone https://github.com/ultralytics/yolov5.git

mv .env.example .env
nano .env  # or with any other editor
# See `🌱 Environment Variables` section for details about the environment variables.

🌱 Environment Variables

Name	Value
TOKEN	Label-Studio Access Token.
LS_HOST	The URL of the label-studio app (e.g., https://label-studio.example.com) – make sure you include https:// in the URL.
DB_CONNECTION_STRING	MongoDB connection string (e.g., mongodb://mongodb0.example.com:27017). See this article for details.
DB_NAME	Name of the main MongoDB database (default: label_studio).
S3_ACCESS_KEY	(Optional) The S3 bucket's Access Key ID.
S3_SECRET_KEY	(Optional) The S3 bucket's Secret Key.
S3_REGION	(Optional) The S3 bucket's region (default: us-east-1).
S3_ENDPOINT	(Optional) The S3 bucket's endpoint/URL server.
EXCLUDE_LABELS	(Optional) Comma-separated list of labels to exclude from processing (e.g., label a,label b).

🗃️ Setup

• To use the GitHub Actions workflows (recommended!), you will need to add every environment variable and its value from .env to the Secrets of your fork (you can find Secrets under Settings).

Click here to show an example of a new secret

🔧 Dataset preparation

• Option 1: Run the JSON to YOLOv5 (data preprocessing) workflow under github Actions.

• Option 2: Run it locally with:

  python birdfsd_yolov5/preprocessing/json2yolov5.py
  mv dataset-YOLO/dataset_config.yml .
  python birdfsd_yolov5/model_utils/relative_to_abs.py

⚡ Training¹

Use the Colab notebook:

📝 Prediction

• Option 1: Run the Predict workflow under github Actions.

• Option 2: Use the Colab notebook:

  

🐳 Using Docker

docker pull alyetama/birdfsd-yolov5:latest

Example Usage

docker run -it --env-file .env alyetama/birdfsd-yolov5 python birdfsd_yolov5/preprocessing/json2yolov5.py

🔖 Related

• BirdFSD-YOLOv5-APP

Footnotes

1. yolov5/wiki/Train-Custom-Data ↩