Skip to content

karstenenglander/EcoLens

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

EcoLens: Waste Classification via Computer Vision

Swift Version Platform License

Overview

EcoLens is an iOS application utilizing machine learning to facilitate sustainable waste management. By leveraging the Apple Neural Engine and the Vision framework, the application performs real-time classification of waste items into three distinct categories: Recycling, Compost, and Landfill. The system is designed with a privacy-centric architecture, performing all inference on-device without external network dependencies for image processing.

Key Features

  • Real-Time Inference: Utilizes a custom CameraController to analyze video frames instantaneously using CoreML.
  • On-Device Processing: Ensures user privacy and offline functionality by processing all data locally on the iOS device.
  • Material Classification: Identifies 12 distinct material classes (including glass, biological, paper, and metal) and maps them to appropriate disposal streams.
  • Photo Library Integration: Allows users to analyze static images imported from the native iOS photo gallery.

Technical Architecture

iOS Application

  • Language: Swift 5
  • UI Framework: SwiftUI
  • Inference Engine: CoreML
  • Camera Handling: AVFoundation (Custom implementation)
  • Concurrency: Grand Central Dispatch (GCD) for non-blocking UI updates

Machine Learning Pipeline

  • Architecture: MobileNetV2 (Transfer Learning)
  • Framework: PyTorch
  • Optimization: Stochastic Gradient Descent (SGD)
  • Model Conversion: coremltools (PyTorch to CoreML conversion with localized probability mapping)
  • Dataset: Standardized Garbage Classification Dataset (15,000+ labeled images)

Repository Structure

This repository follows a monorepo structure containing both the iOS source code and the machine learning development environment.

About

Real-time iOS waste classification using CoreML and MobileNetV2. Features a privacy-centric, offline-first architecture with a custom PyTorch-to-CoreML training pipeline.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors