Face Segmentation

A Streamlit web application that performs semantic segmentation of facial images using the SegFormer model. It allows users to upload an image, detects faces in the image, and performs pixel-wise segmentation to isolate the facial regions from the background.

Features

• Face detection using OpenCV's Haar Cascade classifier
• Semantic segmentation using the pre-trained SegFormer model
• Supports uploading images in JPG, JPEG, and PNG formats
• Displays the original image, detected faces, and segmented image
• Option to override the number of faces limit
• Download the segmented image as a PNG file with transparent background

Installation

Clone the repository:

git clone https://github.com/yourusername/face-segmentation-app.git

Change into the project directory:

cd face-segmentation-app

Install the required dependencies:

pip install -r requirements.txt
pip install -r requirements-backend.txt

Usage

Run the Streamlit app:

streamlit run app_standalone.py

• Open the provided URL in your web browser.
• Upload an image containing a face using the file uploader.
• Click the "Process" button to perform face detection and segmentation.
• View the original image, detected faces, and segmented image.
• Optionally, override the number of faces limit using the sidebar option.
• Download the segmented image by clicking the "Download Segmented Image" button.

Built with

Streamlit, OpenCV, Transformers

Credits

• SegFormer model: HuggingFace, jonathandinu/face-parsing
• Haar Cascade classifier: OpenCV