Skip to main content

​
Introduction

Reciclaje AI uses a specialized YOLOv8 object detection model from Ultralytics, trained specifically for identifying and classifying waste materials and recyclables. The model provides real-time detection capabilities with confidence scoring for accurate waste classification.
The model is available on HuggingFace for easy integration. See the HuggingFace page for access details.

​
Model Specifications

​
Architecture

  • Framework: YOLOv8 (You Only Look Once, version 8)
  • Provider: Ultralytics
  • Model Type: Object Detection
  • Input: RGB images or video frames
  • Output: Bounding boxes with class labels and confidence scores

​
Detection Classes

The model is trained to detect and classify 5 distinct waste categories:

Metal

Class ID: 0Detects metal containers, cans, and metallic waste materials.

Glass

Class ID: 1Identifies glass bottles, jars, and glass fragments.

Plastic

Class ID: 2Recognizes plastic bottles, containers, and plastic packaging.

Carton

Class ID: 3Detects cardboard boxes, paper cartons, and similar materials.

Medical

Class ID: 4Identifies medical waste and healthcare-related disposable items.

​
Confidence Scoring

The model outputs a confidence score for each detection, indicating the certainty of the classification: 
# Confidence values range from 0 to 1
confidence = box.conf[0]  # e.g., 0.85

# Displayed as percentage
confidence_percent = int(confidence * 100)  # e.g., 85%
Higher confidence scores (closer to 100%) indicate greater certainty in the detection. The system uses these scores to filter and validate predictions.

​
Performance Characteristics

​
Real-Time Processing

  • Designed for real-time video stream processing
  • Optimized for webcam input at 1280x720 resolution
  • Low latency detection suitable for interactive applications

​
Accuracy

The model provides reliable predictions for common waste materials when:
  • Objects are clearly visible and well-lit
  • Items are positioned within typical viewing angles
  • Materials match the training data distribution
For information about known limitations and edge cases, see the Limitations page.

​
Use Cases

Reciclaje AI is ideal for:
  • Smart Recycling Bins: Automated waste sorting systems
  • Educational Tools: Teaching proper recycling practices
  • Waste Management: Monitoring and categorizing recyclable materials
  • Environmental Applications: Tracking recycling compliance and behavior

​
Model File

The trained model weights are stored in the best.pt file format, which is the standard PyTorch format used by Ultralytics YOLO: 
from ultralytics import YOLO

# Load the model
model = YOLO('Modelos/best.pt')

# Run inference
results = model(frame, stream=True, verbose=False)

​
Next Steps

Training Data

Learn about the dataset used to train the model

Limitations

Understand model limitations and edge cases

HuggingFace

Access the model on HuggingFace

Quick Start

Get started with Reciclaje AI

Build docs developers (and LLMs) love

Get started for freeTalk to us