Caneca V1 - Client-Side Rubbish Classifier

This repository contains the original proof-of-concept for the Caneca rubbish classifier. It's a web application built with Next.js that performs machine learning inference entirely in the user's browser using ONNX Runtime Web.

The ONNX models are not bundled with the application; they are fetched remotely from a Cloudflare R2 bucket at runtime.

Live Demo: https://caneca-self.vercel.app/

---

Project Status

This application served as a successful prototype. Its current focus is shifting to act as a data collection tool for the next-generation version of the project by integrating with the new V2 backend.

All new backend and future web app development is happening in the V2 repository.

➡️ V2 Repository (Backend & Future Web App): https://github.com/Jotadebeese/caneca_v2

---

Core Technology

• Framework: Next.js
• Language: TypeScript
• ML Inference: ONNX Runtime Web for client-side inference.
• Model Hosting: ONNX models are hosted on and loaded from Cloudflare R2.

Model Conversion

The PyTorch models used in this project were converted to the ONNX format using a custom Python script.

➡️ See the ONNX Exporter Script: https://github.com/Jotadebeese/Rubbish-Classifier-Updated/tree/main/ONNX_exporter

---

Local Development

Prerequisites

• Node.js

Installation

1. Clone the repo:

   git clone [https://github.com/Jotadebeese/caneca.git](https://github.com/Jotadebeese/caneca.git)
   cd caneca

2. Install dependencies (choose one):

   Using npm:

   npm install

   Using yarn:

   yarn

3. Run the dev server (choose one):

   Using npm:

   npm run dev

   Using yarn:

   yarn dev

The application will be available at http://localhost:3000.