Fashion Image Processor

A microservice for generating and managing image embeddings for fashion products using ResNet50. The service provides similarity search capabilities based on visual features extracted from product images.

Features

• Generate embeddings from fashion product images using ResNet50
• Store embeddings in Redis for fast similarity search
• RESTful API for embedding generation and similarity search
• Asynchronous processing with Celery
• Docker support for easy deployment

Prerequisites

• Docker and Docker Compose
• Python 3.8+
• Redis

Quick Start

1. Clone the repository:

git clone <repository-url>
cd fashion-image-processor

2. Create required directories:

mkdir embeddings

3. Start the services using Docker Compose:

docker compose up -d

This will start:

• FastAPI application on port 4000
• Redis on port 6379
• Celery worker for background tasks

API Endpoints

Generate Embeddings for All Products

POST /embeddings/generate

Generates embeddings for all products in the database.

Generate Embedding for a Specific Product

POST /embeddings/product/{product_id}

Generates an embedding only for the product with the given product ID.

Update Embedding for a Specific Product

PUT /embeddings/product/{product_id}

Updates the embedding for the product with the given product ID.

Delete Embedding for a Specific Product

DELETE /embeddings/product/{product_id}

Deletes the embedding for the product with the given product ID.

Find Similar Products

GET /similar-products/{product_id}?top_n=5

Returns the most similar products based on image embeddings.

Health Check

GET /ping

Returns the service status.

Development

1. Create a virtual environment:

python -m venv .venv
source .venv/bin/activate  # Linux/Mac
# or
.venv\Scripts\activate  # Windows

2. Install dependencies:

pip install -r requirements.txt

3. Run Celery worker:

celery -A app.celery_config.celery_app worker --loglevel=info

Architecture

The service consists of several components:

• FastAPI application for HTTP endpoints
• ResNet50 model for feature extraction
• Redis for embedding storage
• Celery for background task processing

Configuration

Environment variables:

• API_BASE_URL: Base URL for the fashion API
• REDIS_URL: Redis connection URL
• CELERY_BROKER_URL: Celery broker URL

API Usage Examples

Generate All Embeddings

curl -X POST "http://localhost:4000/embeddings/generate" \
  -H "Content-Type: application/json"

Generate Embedding for a Specific Product

curl -X POST "http://localhost:4000/embeddings/product/123" \
  -H "Content-Type: application/json"

Update Embedding for a Specific Product

curl -X PUT "http://localhost:4000/embeddings/product/123" \
  -H "Content-Type: application/json"

Delete Embedding for a Specific Product

curl -X DELETE "http://localhost:4000/embeddings/product/123" \
  -H "Content-Type: application/json"

Find Similar Products

curl -X GET "http://localhost:4000/similar-products/123?top_n=5" \
  -H "Content-Type: application/json"

License

MIT License

Create the client

client = ApiClient(base_url="https://fashion.aknevrnky.dev")

Get image bytes by product ID

image_data = await client.get_product_image(product_id=1)

Or directly from a URL

image_data = await client.get_product_image(image_url="https://fashion.aknevrnky.dev/storage/products/10000.jpg")

Process with PIL/Pillow

from PIL import Image import io image = Image.open(io.BytesIO(image_data))

Now the image can be processed for training