/VGG-Face-Avatar-Generation

Generates unique 3D avatars using VGG-Face CNN

Primary LanguageLua

Avatar VGG Face Generation from Photos

Reference: https://patniemeyer.github.io/2018/10/29/generating-memoji-from-photos.html

eval-memoji.lua

This script grabs screenshots from a specified region of the screen, scores them with VGG Face against a folder of reference photos, and saves each captured image to a folder naming the file with its score. The script waits for you to hit enter on the keyboard for each capture.

compare-memoji-multi.lua

This script takes a folder of references and a folder of target photos and uses VGG Face to rank the targets based on the output of a selected layer of the network and a simple dot product similarity measure. The top three matching images are displayed along with their scores.

If you have a folder of ranked output from eval-memoji.lua representing e.g. a Memoji with each feature variation, you can run this script on that output folder to match new reference files or experiment with other similarity metrics.

Installing on Mac OS

Homebrew

Install the Homebrew package manager:

https://brew.sh/

Torch

Clone the torch repo to a folder called distro. (Some of the install scripts insist on finding this name in the path.)

git clone https://github.com/torch/torch7.git ./distro

Run the install-deps script.

If anything important reports as outdated you can use brew upgrade package to update it.

Run the install script.

Install loadcaffe for loading models:

luarocks install loadcaffe

If you have trouble with qlua insisting on version 4 of Qt you can switch your installed version to 4 with brew :

brew switch qt 4.8.7_2