README.md

metadata

license: mit
language:
  - ko
tags:
  - image-classification
  - computer-vision
  - korean
  - secondhand-market
  - e-commerce
  - ensemble
  - tensorflow
  - keras
datasets:
  - custom
metrics:
  - accuracy
model_type: ensemble
pipeline_tag: image-classification
widget:
  - src: https://example.com/sample_image.jpg
    example_title: Sample Product Image
base_model:
  - google/efficientnet-b0
  - microsoft/resnet-50
  - facebook/deit-base-distilled-patch16-224
library_name: tensorflow

Korean Secondhand Market Classifier

Model Overview

AI model for automatic categorization of Korean secondhand market product images. 70.61% accuracy achieved with 7-model ensemble system.

Supported Categories

1. 가구 (Furniture) - beds, sofas, desks, chairs
2. 생활용품 (Household items) - kitchenware, cleaning supplies, storage
3. 전자기기_도서 (Electronics/Books) - smartphones, laptops, books, e-books
4. 취미_게임 (Hobbies/Games) - game consoles, board games, sports equipment
5. 패션_뷰티 (Fashion/Beauty) - clothing, shoes, cosmetics, accessories

Performance

• Ensemble Accuracy: 70.61%
• Individual Models: 7 models (EfficientNet, ResNet50V2, DenseNet, etc.)
• Input Size: 224x224 RGB images

Usage

# Install dependencies
pip install fastapi uvicorn tensorflow pillow huggingface_hub

# Download and run
from huggingface_hub import snapshot_download
repo_path = snapshot_download("bihan3876/my_model")

# Run API server
import subprocess
subprocess.run(["python", f"{repo_path}/api_server.py"])

File Structure

models/
├── ensemble/              # Ensemble models (349MB)
│   ├── EfficientNetB0_best.keras
│   ├── ResNet50V2_best.keras
│   └── ... (7 models)
└── serving/               # Serving models
    ├── model_optimized.tflite  # 24MB
    └── TensorFlowLiteInferenceService.java

License

MIT License