(ACMMM 2021 Oral) SfM Face Reconstruction Based on Massive Landmark Bundle Adjustment

This repository shows two tasks: Face landmark detection and Face 3D reconstruction, which is described in this paper: Deep Unsupervised 3D SfM Face Reconstruction Based on Massive Landmark Bundle Adjustment.

Installation

  1. Clone the repository.
  2. install dependencies.
pip install -r requirement.txt

Face landmark detection

RedAndGreen

Picture1_crop

Running a pre-trained model

  1. Download landmark pre-trained model at GoogleDrive, and put it into FaceLandmark/model/
  2. Run the test file
python Facial_landmark.py

Face 3D reconstruction

Stirling-ESRC-3D

Facescape-face

Running a pre-trained model

  1. Download face 3D reconstruction pre-trained model at GoogleDrive, and put it into FaceReconstruction/checkpoints/

  2. Run the inference.py file to generate disparity map

python inference.py --dataset-dir './FaceReconstruction/test_image/' --output-dir './FaceReconstruction/output/' --pretrained './FaceReconstruction/checkpoints/dispnet_model_best.pth.tar' --resnet-layers 18 --output-disp
  1. Run the generate_ply.py file to generate point cloud .ply file
python generate_ply.py

GitHub

https://github.com/BoomStarcuc/3DSfMFaceReconstruction