📊 AWS Pinterest Image Analytics Pipeline

End-to-End Computer Vision & Cloud Integration Project

---

🎯 Project Overview

This project implements a complete end-to-end image analytics pipeline using:

• AWS SageMaker
• AWS Rekognition
• ImageAI (YOLOv3)
• Google Firebase Firestore
• Python (Pandas, Requests, Boto3)

The objective was to scrape Pinterest images, analyze them using both cloud-based and local computer vision models, compare the outputs, and store structured results in a scalable cloud database.

This project demonstrates cloud AI integration, object detection, data engineering, and business analytics thinking.

---

📁 Dataset Description

The dataset consists of:

• 75 scraped Pinterest images
• Metadata including:
  • Pin URL
  • Image URL
  • Board/topic
  • Engagement metrics
  • Notes on visual appeal

Primary dataset file:

• pinterest_scraped_dataset.xlsx

A gallery preview of scraped images:

📸 Sample Scraped Images

The dataset mainly includes indoor home décor images featuring plants, sofas, books, warm lighting, and furniture arrangements.

---

1️⃣ AWS Rekognition – Image Labeling (Q3)

Approach

• Selected 20 images from the scraped dataset
• Downloaded image bytes dynamically
• Used AWS Rekognition detect_labels() API
• Extracted:
  • Top label
  • Confidence score
  • Labels above 90% confidence threshold
• Exported structured CSV results

Output file:

• aws_rekognition_image_labels.csv

Key Characteristics

• Cloud-based
• Fast and scalable
• Returns scene-level labels
• No bounding boxes (in this implementation)
• Structured and clean output

Example Output

• Architecture – 100%
• Indoors – 99.99%
• Home Decor – 99.99%

---

2️⃣ ImageAI (YOLOv3) – Object Detection (Q4)

Approach

• Processed all 75 images
• Used YOLOv3 model weights
• Detected multiple objects per image
• Extracted:
  • Object name
  • Confidence score
  • Bounding box coordinates
• Saved annotated images
• Exported structured CSV with 363 detections

Output file:

• yolov3_object_detection_results.csv

Top Detected Objects

• Potted plants
• Books
• Sofas
• Chairs
• Vases

This aligns strongly with the indoor décor theme of the dataset.

Strengths

• Multiple object detection
• Bounding box localization
• Rich visual output
• High interpretability

---

3️⃣ Model Comparison (Q5)

Similarities

• Both methods classify image content
• Both export structured CSV results
• Both provide confidence scores

Differences

AWS Rekognition	ImageAI (YOLOv3)
Cloud-based	Local processing
Scene-level labeling	Object-level detection
Faster setup	More detailed output
No bounding boxes	Bounding boxes included
Scales easily	Slower and resource-heavy

When to Use Which

• Use Rekognition for large-scale production classification
• Use ImageAI when detailed object positioning is required
• Use GenAI tools for natural-language scene description (not structured analytics)

---

4️⃣ Firebase Firestore Integration (Q7)

All datasets were uploaded to Google Firebase Firestore:

Collections created:

• q1_pins
• q2_followers
• q3_rekognition
• q4_imageai
• q6_color_style

Each row of analysis results was stored as a document in Firestore.

Benefits

• Cloud storage scalability
• Real-time querying
• Structured document model
• Easier future integration with web applications

---

5️⃣ Business Recommendations (Q8)

Based on visual analytics and engagement patterns:

Findings

• Indoor décor images with plants and warm lighting drive strong engagement
• Multiple objects in cozy layouts correlate with saves and reactions
• Visual consistency matters

Recommendation

Pinterest could implement:

• AI-powered “See in My Room” feature
• AR-based object placement
• Palette-aware product recommendations
• Video-first interactive discovery flow

This would convert inspiration into purchase with higher engagement and reduced friction.

---

📊 Technologies Used

• Python
• Pandas
• Boto3
• AWS Rekognition
• AWS SageMaker
• ImageAI (YOLOv3)
• Google Firebase Firestore
• Requests
• PIL
• JSON processing

---

🧠 Technical Highlights

• Dynamic image downloading
• Cloud API integration
• Confidence threshold filtering
• Bounding box extraction
• One-row-per-object structured CSV generation
• Batch upload to Firestore
• Scalable cloud pipeline design

---

📂 Repository Structure

01_AWS_Rekognition_Image_Labeling.ipynb
02_ImageAI_Object_Detection_and_Firebase_Pipeline.ipynb

pinterest_scraped_dataset.xlsx
aws_rekognition_image_labels.csv
yolov3_object_detection_results.csv
image_color_style_analysis.csv

pinterest_image_gallery_preview.png
README.md

---

👤 Author

Nicky Kumari
Cloud Analytics | Computer Vision | Data Engineering | AI Integration