Clothing segmentation with Hugging Face

Python, Computer Vision, Image Segmentation, Hugging Face, REST API, Deep Learning, Data Science, NumPy, Matplotlib, Pillow

Fashion Trend Intelligence – Clothing Segmentation with Hugging Face

Context

As part of a demonstration project, I developed a prototype to evaluate the use of a computer vision model hosted on Hugging Face Serverless endpoints to automatically identify clothing present in images. The objective was to validate the technical feasibility before a larger-scale deployment for trend analysis in the fashion industry.

Objectives

• Integrate an image segmentation model via a REST API.
• Automatically segment different clothing items.
• Generate visualizations of the results.
• Evaluate model performance.
• Estimate the operating cost of the solution at scale.

Architecture

Images
   │
   ▼
Script Python
   │
   ▼
API Hugging Face Inference
(segformer_b3_clothes)
   │
   ▼
Masques de segmentation
   │
   ▼
Visualisation + Évaluation

Technologies Used

• Python
• Requests
• Pillow (PIL)
• NumPy
• Matplotlib
• python-dotenv
• API Hugging Face Inference
• Git
• Visual Studio Code

Achievements

AI Service Integration

• Secure authentication via API token.
• Environment variable management (.env).
• Communication with a Serverless AI API.
• Automatic image processing.

Image Segmentation

• Sending images to the SegFormer Clothes model.
• Retrieval of segmentation masks.
• Association of detected classes with clothing items.
• Generation of visualizations (original image / segmentation).

Model Evaluation

Implementation of an evaluation approach including:

• dataset separation

  • training
  • validation
  • test

• definition of performance indicators such as:

  • Intersection over Union (IoU)
  • Dice Score
  • F1 Score

• qualitative analysis of the obtained results.

Cost Study

Evaluation of the cost of using Hugging Face Serverless endpoints for a production scenario of 500,000 images over 30 days, to estimate the economic viability of the solution.

Demonstrated Skills

Artificial Intelligence

• Use of a pre-trained model
• Computer Vision
• Semantic Segmentation
• AI API Consumption

Python Development

• Image manipulation
• HTTP Requests
• Error handling
• Environment variables
• Python project organization

Data Science

• Model evaluation
• Selection of appropriate metrics
• Performance analysis
• Experimental validation

Cloud & API

• Use of a Serverless AI service
• Token authentication
• Estimation of operating costs

Outcome

This project allowed me to put into practice the integration of a ready-to-use AI service to quickly build a functional computer vision prototype, while applying a comprehensive approach ranging from API calls to performance evaluation and deployment cost analysis.

Acquired Skills

• Python
• Computer Vision
• Image Segmentation
• Hugging Face Inference API
• REST API
• Requests
• Pillow
• Matplotlib
• NumPy
• python-dotenv
• AI Model Evaluation
• IoU
• Dice Score
• F1 Score
• Data Visualization
• Git
• Secret Management
• Cloud Cost Analysis