Skip to content

Latest commit

 

History

History
71 lines (53 loc) · 3.22 KB

README.md

File metadata and controls

71 lines (53 loc) · 3.22 KB
Raw

MVL Face Datasets

About the Project

MVL Face Datasets is a collection of image processing pipelines written in MATLAB that generate variations of well-known face datasets. The current collection includes following pipelines:

Dependencies

All pipelines rely on MatConvNet 1.0-beta25 and OpenFace 2.2.0 for facial landmark detection. CFD-cutout pipeline uses SHINE toolbox to match lightness histograms across images.

Getting Started

Before running each pipeline, please run the following script to download and set up required libraries (MatConvNet, OpenFace, and SHINE toolbox):

>> Step0_Libraries

Pipelines can be found under the pipelines/ directory.

KDEF-masked

To generate masked face images from KDEF:

  1. Download KDEF_and_AKDEF.zip file from https://www.kdef.se and unpack it.
  2. Copy KDEF/ directory and paste it under pipelines/KDEF-masked/ directory.
  3. Change the MATLAB working directory to the pipelines/KDEF-masked/ directory.
  4. Run the following scripts:
>> Step1_PrepareImages
>> Step2_DetectLandmarks
>> Step3_BuildDataset
  1. Please check imgs-dataset/ directory for masked face images, dataset.xlsx and dataset.mat files for the map of image names for each model × expression × angle.

CFD-masked

To generate masked face images from CFD:

  1. Download cfd.zip file from https://www.chicagofaces.org and unpack it.
  2. Copy CFD/ directory (found under CFD Version 3.0/Images/) directory and paste it under pipelines/CFD-masked/ directory.
  3. Change the MATLAB working directory to the pipelines/CFD-masked/ directory.
  4. Run the following scripts:
>> Step1_PrepareImages
>> Step2_DetectLandmarks
>> Step3_BuildDataset
  1. Please check imgs-dataset/ directory for masked face images, dataset.xlsx and dataset.mat files for the list of image names for each model.

CFD-cutout

To generate face-cutout images from CFD:

  1. Download cfd.zip file from https://www.chicagofaces.org and unpack it.
  2. Copy CFD/ directory (found under CFD Version 3.0/Images/) directory and paste it under pipelines/CFD-cutout/ directory.
  3. Change the MATLAB working directory to the pipelines/CFD-cutout/ directory.
  4. Run the following scripts:
>> Step1_PrepareImages
>> Step2_DetectLandmarks
>> Step3_CutoutImages
>> Step4_BuildDataset
  1. Please check imgs-dataset/ directory for face-cutout images, dataset.xlsx and dataset.mat files for the list of image names for each model.

Acknowledgment

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (Ministry of Science and ICT) (No. 2022R1C1C1008628).