SOTC_CloudVision

Sigils of The Codex

This project demonstrates how to use Google’s Cloud Vision API to identify recyclable items in images using Python and Jupyter Notebook. The notebook guides you through running inferences, retrieving annotation labels, mapping them to recyclable categories, and building a scoring system to determine the best prediction for each image.

Features

• Object Localization: Detects objects in images using Google Cloud Vision API.
• Category Mapping: Maps detected objects to recyclable categories defined in config.json.
• Scoring System: Resolves ties and improves accuracy using label detection.
• Config Automation: Automatically expands the label database for better recognition.
• Image Preprocessing: Optimizes images for inference.

Getting Started

Prerequisites

• Python 3.7+
• Google Cloud account and service account key (serviceAccountKey.json)
• Jupyter Notebook

Installation

1. Clone this repository and navigate to the project folder.
2. Install required packages:

   pip install google-cloud-vision pillow
   

3. Place your Google Cloud service account key as serviceAccountKey.json in the project root.

Usage

1. Open SOTC_MasterCopy.ipynb in Jupyter Notebook.
2. Follow the notebook cells to:
   • Authenticate with Google Cloud Vision API
   • Load and preprocess images
   • Run object localization and label detection
   • Map results to recyclable categories
   • Expand the label database for improved accuracy

Folder Structure

images/                # Sample images for testing
config.json            # Maps categories to annotation labels
serviceAccountKey.json # Google Cloud credentials
SOTC_MasterCopy.ipynb  # Main notebook

How It Works

• The notebook walks you through uploading images, running inferences, and updating the label database.
• You can add new recyclable categories and expand their label sets by running images through the helper functions.
• The pipeline improves reliability as more images and labels are added.