How mwasifanwar/oceanmind works
Project Type and Purpose
• oceanmind is a domain-specific AI system (full-stack machine learning application) that processes underwater imagery and video to:
• Track marine life (fish_detector)
• Monitor coral reef health (coral_reef_detector)
• Detect illegal fishing (illegal_fishing_detector)
• It provides both a training pipeline (train.py) and an inference/service pipeline (main.py + api/app.py).
File-Structure Analysis
• Top-level directories:
• api/ → REST API service (app.py)
• config/ → application configuration (settings.py)
• data/ → data ingestion & loaders (dataloader.py)
• models/ → model definitions & inference wrappers
• processing/ → video/image processing pipeline
• utils/ → shared utilities (image processing, visualization)
• tests/ → unit tests
• Root scripts:
• main.py → orchestrates end-to-end inference pipeline
• train.py → orchestrates model training
• Requirements.txt pins dependencies.
Key Components
a. Data Sources
• Raw underwater imagery/videos (on-disk or cloud storage)
b. Data Ingestion Layer
• data/dataloader.py
c. Preprocessing Layer
• utils/image_processing.py
• processing/video_processor.py
d. Inference Engine
• models/*.py (fish, coral reef, illegal fishing detectors)
e. Post-processing & Visualization
• utils/visualization.py
f. API Layer
• api/app.py (Flask/FastAPI service)
g. Orchestration & CLI
• main.py (invokes ingestion → preprocessing → inference → output)
h. Configuration
• config/settings.py
i. Training Pipeline
• train.py (uses data loader, processing, models to train new weights)
j. Testing
• tests/test_models.py
Interactions & Data Flow
• (1) Ingestion: Raw data → Data Loader
• (2) Preprocessing: Video Processor + Image Processing
• (3) Inference: Preprocessed frames → Detector Models
• (4) Post-processing: Inference results → Visualization outputs
• (5) API: External clients → API Layer → calls main.py logic → returns detection results/visuals
• (6) Training: Raw data → train.py → model definitions → new model artifact
• (7) Configuration: settings injected at startup (both for API and training)
Architectural Patterns & Principles
• Layered architecture separating data, processing, inference, API
• Modular, plug-and-play detectors (each model in its own module)
• Single responsibility for each script/module
• Configuration-driven behavior (via config/settings.py)
• Reusable utilities for image processing and visualization
Inclusion Checklist
• Clear labels for every component (module name + file path)
• Distinction between training and inference pipelines
• Configuration source (config/settings.py) pointed to by all runtime components