Skip to content

Latest commit

 

History

History
92 lines (83 loc) · 4.88 KB

README.md

File metadata and controls

92 lines (83 loc) · 4.88 KB
Raw

FaceSwap

Overview

PHASE 1: TRADITIONAL APPROACH

alt test

1. Facial Landmarks Detection

To perform Face swap, we need to first detect a human face in a given image frame. In order to detect a human face, we utilise a Histogram of Oriented Gradients (hog)

  • Support Vector machine (svm) based face detector, that is built in the dlib library alt text

2. Face Warping

Using Triangulation

alt text

Using Thin Plate Spline

alt text

3. Blending

The source face, once reshaped using one of the above warping methods, were blended in the location of the destination image frame using Poisson blending. To perform Poisson blending, we need compute a mask of the face in destination image, where the region of interest (ROI) of the face has white pixels and the rest of the frame is black. alt text

PHASE 2: DEEP LEARNING APPROACH

In this section, we implement a deep learning network proposed by Feng et al., in a research work titled ”Joint 3D Face Reconstruction and Dense Alignment with Position Map Regression Network”. In this research, they compute a 2D representation of the face called UV position map, which records the 3D shape of a complete face in UV space, then train a simple Convolutional Neural Network to regress it from a single 2D image. We utilised code from github provided by the authors for our Face Swap pipeline.

Results

Original Video

Original video

1. delTri

Original video

2. TPS

Original video

3.PRNet

Original video

Run the code

  1. Download the Face Landmarks detector file from here, unzip and copy it to Misc/ Folder
  2. Download the PRNet pretrained model file from here, unzip and copy it to phase2/Data/net-data/ Folder
  3. Copy the dlib facial marker detector model to Code/phase1/Misc/
  4. For PRNet, copy the pretrained model file to Code/phase2/Data/net-data/
  5. To run the program
    cd Code/.

1. delTri

Swap face from image and video:

python3 Wrapper.py --method delTri --DataPath ../Data/TestSet/ --VideoFileName Test1.mp4 --RefImageFileName Rambo.jpg --SavePath Test1OutputTri.mp4 --PredictorPath ./phase1/Misc/shape_predictor_68_face_landmarks.dat --use_filter 1

python3 Wrapper.py --method delTri --DataPath ../Data/TestSet/ --VideoFileName Test3.mp4 --RefImageFileName Scarlett.jpg --SavePath Test3OutputTri.mp4 --PredictorPath ./phase1/Misc/shape_predictor_68_face_landmarks.dat --use_filter 0

Swap faces within a video:

python3 Wrapper.py --method delTri --DataPath ../Data/TestSet/ --VideoFileName Test2.mp4 --SavePath Test2OutputTri.mp4

2. TPS

Swap face from image and video:

python3 Wrapper.py --method TPS --DataPath ../Data/TestSet/ --VideoFileName Test1.mp4 --RefImageFileName Rambo.jpg --SavePath Test1OutputTPS.mp4 --PredictorPath ./phase1/Misc/shape_predictor_68_face_landmarks.dat --use_filter 1

python3 Wrapper.py --method TPS --DataPath ../Data/TestSet/ --VideoFileName Test3.mp4 --RefImageFileName Scarlett.jpg --SavePath Test3OutputTPS.mp4 --PredictorPath ./phase1/Misc/shape_predictor_68_face_landmarks.dat --use_filter 0

Swap faces within a video:

python3 Wrapper.py --method TPS --DataPath ../Data/TestSet/ --VideoFileName Test2.mp4  --SavePath Test2OutputTPS.mp4

3.PRNet

Swap face from image and video:

python3 Wrapper.py --method PRNet --DataPath ../Data/TestSet/ --VideoFileName Test1.mp4 --RefImageFileName Rambo.jpg --SavePath Test1OutputPRNet.mp4 --PredictorPath ./phase1/Misc/shape_predictor_68_face_landmarks.dat --use_filter 1

python3 Wrapper.py --method PRNet --DataPath ../Data/TestSet/ --VideoFileName Test3.mp4 --RefImageFileName Scarlett.jpg --SavePath Test3OutputPRNet.mp4 --PredictorPath ./phase1/Misc/shape_predictor_68_face_landmarks.dat --use_filter 0

Swap faces within a video:

python3 Wrapper.py --method PRNet --DataPath ../Data/TestSet/ --VideoFileName Test2.mp4  --SavePath Test2OutputPRNet.mp4