Face Recognition Plugin for Dataiku DSS

Overview

The Face Recognition Plugin for Dataiku DSS provides tools to automate the process of face recognition within your data workflows. This plugin includes two main recipes:

1. Face Encoding Recipe: Converts a managed folder containing images or a dataset containing URLs toward images into face encodings. Encodings are a 128 floats vectors, and one encoding is produced for each face present in the source pictures.
2. Reference Dataset Builder Recipe: Builds a reference dataset containing one grouping GUID for every matching face encoding.

Installation

To install the Face Recognition Plugin, follow these steps:

1. Make sure cmake is installed on your DSS instance. On a Mac, run brew install cmake in a terminal. More details on the necessary installation steps here.
2. On Dataiku DSS, go to App > Plugins > Add Plugin > Fetch from Git repository > set [email protected]:alexbourret/dss-plugin-face-recognition.git in repository URL
3. Follow the installation prompts to complete the setup.

Recipes

Face Encoding Recipe

The Face Encoding Recipe transforms images into face encodings. This recipe can process:

• A managed folder containing images.
• A dataset containing URLs toward images.

Inputs

• Managed Folder: A folder containing image files.
• Dataset with URLs: A dataset where each row contains a URL pointing to an image.

Outputs

• Face Encodings Dataset: A dataset containing the face encodings extracted from the input images.

Configuration

• URLs column: If the recipe's input is a dataset, specify the name of the column containing the URLs towards the images containing the faces to encode.

Reference Dataset Builder Recipe

The Reference Dataset Builder Recipe creates a reference dataset containing one grouping GUID for every matching encoding (faces). This recipe helps in identifying and grouping similar faces.

Inputs

• Face Encodings Dataset: The output from the Face Encoding Recipe or any dataset containing face encodings.

Outputs

• Reference Dataset: A dataset with a unique grouping GUID for each set of matching face encodings.

Configuration

• Uknown encodings column: Select the column containing the encodings in the dataset of faces not yet referenced
• Known encodings column: Select the column containing the encodings in the dataset of already referenced faces
• Known references column: Select the column containing the reference GUID

Usage

Step-by-Step Guide

1. Create a Managed Folder or Dataset:

   • Upload images to a managed folder in Dataiku DSS.
   • Alternatively, create a dataset containing URLs toward images.

2. Run the Face Encoding Recipe:

   • Create a new recipe and select "Face Encoding Recipe".
   • Configure the input (managed folder or dataset with URLs) and output (face encodings dataset).
   • Run the recipe to generate face encodings.

3. Build the Reference Dataset:

   • Create a new recipe and select "Reference Dataset Builder Recipe".
   • Configure the input (face encodings dataset) and output (reference dataset).
   • Run the recipe to build the reference dataset with grouping GUIDs.

Example Workflow

1. Upload Images: Upload a set of images to a managed folder in Dataiku DSS.
2. Encode Faces: Use the Face Encoding Recipe to convert these images into face encodings.
3. Build Reference Dataset: Use the Reference Dataset Builder Recipe to create a reference dataset with grouping GUIDs for matching faces.

License

This plugin is distributed under the Apache License version 2.0