sanjeevongit / image-identification

Label and Classifying Images

Image-Identification

Comparing AWS, Azure, and Google Cloud Vision

Comparison of face detection, text detection, and object tagging between three cloud service providers.

Units of comparison:

• Accuracy
• Cost
• Speed/Latency

---

Link to Paper: In Progress

COEN 241, Cloud Computing

Santa Clara University

---

Google Results:

• Labeler:
  • Speed: 70.8s for 250 images
  • Mean Accuracy: 58.1%
• Text Extractor:
  • Speed: 160.7s for 175 images
  • Mean Jaccard Similarity: 45.7%
• Face Detector:
  • Speed: 303.6s for 550 images
  • Mean Accuracy: 99.8%

Azure Results:

• Cost

  • Free Tier: 5000 transactions free/month, capped at 200$ credit.
  • Paid Tier:

    • Tag, Face:

      • 0-1M transactions: $1 per 1,000 transactions
      • 1M-5M transactions — $0.80 per 1,000 transactions
      • 5M-10M transactions — $0.65 per 1,000 transactions
      • 10M-100M transactions — $0.65 per 1,000 transactions
      • 100M+ transactions — $0.65 per 1,000 transactions

    • OCR:

      • 0-1M transactions — $1.50 per 1,000 transactions
      • 1M-5M transactions — $1 per 1,000 transactions
      • 5M-10M transactions — $0.65 per 1,000 transactions
      • 10M-100M transactions — $0.65 per 1,000 transactions
      • 100M+ transactions — $0.65 per 1,000 transactions

• Labeler:

  • Speed: 108.9 s for 250 images
  • 56% accuracy within Top-5 tags
  • 46.4% accuracy with confidence > 50%

• Text Extractor:

  • Speed: 89.3s for 175 images
  • Mean Jaccard Similarity: 23.4%

• Face Detection:

  • Speed: 192.9s for 550 images
  • Mean Accuracy: 99.8%
  • Confidence score only available for certain facial features, not detection

More can be located here.

AWS Results:

• Cost
  • FreeTier: 5000 images/month
  • PaidTier:
    • $1 per 1000 images for first 1Million images
    • $.80 per 1000 images for next 9Million images
    • $.60 per 1000 images for next 90Million images
    • $.40 per 1000 images if over 100Million images
• Labeler:
  • Speed: 230s for 250 images (using S3 bucket)
  • Speed: 351s for 250 images (using direct upload)
  • Mean Accuracy: 58.0%
  • Mean Accuracy: 49.1% (Raw without lemmatize)
  • Mean Accurancy (Synonyms): 78.6%
• Text Extractor:
  • Speed: 356s for 175 images (using S3 bucket)
  • Speed: 592s for 175 images (using direct upload)
  • Mean Jaccard Similarity: 48.5%
• Face Detector:
  • Speed: 645s for 550 images (using S3 bucket)
  • Speed: 957s for 550 images (using direct upload)
  • Mean Accuracy: 96.3%