Database Schema Overview

This page outlines the database schema designed to support the core components of the wildfire detection pipeline. The schema is structured for scalability and maintainability, with a focus on enabling post-handoff use by the partner team.

---

images Table

Tracks all image assets in the pipeline.

Field	Type	Description
id	String (UUID)	Unique image identifier
image_path	String	Local path to the image file
label_path	String	Local path to the corresponding label JSON file
classes	JSON	Array of class labels (e.g., ["Fire", "Smoke"])
bboxes	JSON	Array of bounding boxes ([[x1, y1, x2, y2], ...])
label_status	Enum	One of auto, human, or mismatch
semi_supervised	JSON	Additional info from pre-labeling (if any)
tags	JSON	Array of tags (e.g., ["augmented", "smoke"])

---

models Table

Stores metadata for each trained, distilled, or quantized model.

Field	Type	Description
id	String (UUID)	Model identifier
model_path	String	Path to the model weights file
model_type	Enum	One of base, distilled, or quantized
model_status	Enum	training or trained
train_image_ids	JSON	UUIDs of images used for training
val_image_ids	JSON	UUIDs of images used for validation
test_image_ids	JSON	UUIDs of images used for testing
training_history	JSON	Training metrics, loss, and other metadata

---

experiments Table

Tracks experiment configurations and evaluation metrics.

Field	Type	Description
id	String (UUID)	Unique experiment identifier
model_id	String	Foreign key to models.id
augmentation	JSON	Augmentation settings (e.g., {"flip": true})
random_seed	String	Seed used for reproducibility
performance	JSON	Model performance metrics (e.g., {"mAP": 0.89})
config	JSON	Training configuration (e.g., {"batch_size": 16})

---

.env Configuration

Set up your database connection in a .env file at the root of the project:

# Default SQLite setup
DATABASE_URL=sqlite:///db.sqlite3

# Optional PostgreSQL setup
# DATABASE_URL=postgresql://username:password@localhost:5432/your_database