madhavkarthi/2025-24679-HW1-Images-mkarthik

🥄 2025-24679-HW1-Images (Cutlery Classification)

📌 Purpose

This dataset was created for an academic assignment on data collection and augmentation. It is intended to support binary image classification tasks: detecting whether an image contains cutlery or not.

📊 Composition

• —Total size (original): 32 images
• —21 containing cutlery
• —11 without cutlery
• —Augmented size: ~352 images (32 originals + 320 synthetic variants)
• —Image size: 224×224 pixels (resized and cropped to square)
• —File format: PNG/JPG (depending on source images)

📥 Collection

• —Images were collected manually and stored in Google Drive folders:
• —Cutlery
• —No Cutlery
• —Each image was resized, square-cropped, and converted to RGB.
• —No personally identifiable information (PII) or sensitive content was collected.

🔧 Preprocessing & Augmentation

Preprocessing

• —Cropped to square
• —Resized to 224×224
• —Converted to RGB

Augmentation

• —Geometric: random resized crop, rotations, flips
• —Photometric: color jitter, autocontrast, sharpness adjustment
• —Occlusion: random erasing (cutout-like)
• —RandAugment (optional) for stronger perturbations

🏷️ Labels

• —Binary classification:
• —0 → No Cutlery
• —1 → Cutlery
• —Stored as a ClassLabel feature: ["No Cutlery", "Cutlery"]

📂 Splits

• —original: 32 manually collected images
• —augmented: ~320 synthetic images (10 augmentations per original)

🎯 Intended Use & Limits

• —Use: coursework, experimentation with image classification models
• —Not for: production deployment or real-world cutlery detection
• —Models trained on this dataset will not generalize due to its small size

⚖️ Ethical Notes

• —Dataset avoids sensitive content (no faces, no private data).
• —No known safety concerns.
• —Synthetic augmentation does not fully capture real-world variability.

📜 License

• —CC-BY-4.0 (Creative Commons Attribution 4.0 International)
• —Free to use for educational and research purposes with attribution.

🤖 AI Usage Disclosure

• —AI tools (ChatGPT + torchvision augmentations) assisted in designing the preprocessing, augmentation, and documentation pipeline.
• —Original images were collected manually, not AI-generated.