README.md

##To use the virtual python environment in "one-shot-envs", do the following: conda create --prefix -c menpo python=3.5.4 -y cp -rf ./lsfm-envs conda activate conda install -c menpo cyvlfeat suitesparse scikit-sparse cyrasterize conda install -c anaconda tensorflow-gpu tensorflow-estimator flask pip install menpo menpofit menpodetect menpo3d dlib lsfm eos-py python-pcl

... go to the folder of virtual environment/lib/python3.5/site-packages/menpo3d edit extractimage.py --> --original-- from menpo3d.rasterize import GLRasterizer, model_to_clip_transform --new-- from menpo3d.rasterize.opengl import GLRasterizer from menpo3d.rasterize import model_to_clip_transform

Constructing a new model from a private dataset

Step-by-step Instructions

Put the input models and landmark files into a folder, e.g. /home/u/workspace/data/Unre_3DMM_Obj/Landmark_68

There are two python scripts in the directory:

preprocessor.py: process the obj files and landmark and copy into lsfm working directory

clean.py: clear all pre-processored result

cp -rf [folder containing model and landmark files] /home/u/workspace/data/Unre_3DMM_Obj/Landmark_68 cd /home/u/workspace/data/Unre_3DMM_Obj/Landmark_68 (optional)python3 clean.py python3 preprocessor.py /home/u/workspace/lsfm/input_dir/

activate lsfm virtual environment

go to lsfm directory and start the training

conda activate lsfm cd /home/u/workspace/lsfm (optional)cd ./output_dir; ./clean.sh; cd .. cd ./bin python3 lsfm -i ../input_dir/ -o ../output_dir/ -j 11

the result being used for fitting is stored in /home/u/workspace/lsfm/output_dir/pca_result

##if tempalte needs to be initialized, use the following command python3 lsfm init /home/u/miniconda2/envs/lsfm/lib/python3.5/site-packages/lsfm/data/template0.ply python3 lsfm init /home/u/workspace/practice/3DMM/PublicMM1/01_MorphableModel.mat

=======

3D-Morphable-Model-training

This program is to train the face 3D morphable model (3DMM)

f3b89ce0429047300bf0967a43d7b995b8ca6e2a