The Face and Object Recognition Module for Co-Creation Space contains neural networks capable of generating automatic tags based on face and object recognition for content filtering in the Co-Creation Space.
Follow the instructions at https://pytorch.org/ to install PyTorch for your
system using either conda or pip. If you have a CUDA compatible graphics
card, ensure to install a CUDA compatible version of PyTorch to allow the
deep models to run on the GPU.
To install the facerec package and facetool console script, run:
pip install .
This will install the facetool script on your environment path.
The facerecdb.py file integrates the face recognition module to Co-Creation Space. Once Co-Creation Space is running with AWS credentials set up in the AWS CLI, simply run the facerecdb.py file. Make sure database credentials are correct.
The module will add tags based on faces found to videos in the Co-Creation Space. For faces to be correctly tagged, upload to Co-Creation Space (i.e. create a post) at least one image of the faces of the individuals that appears in videos. Name the post with the name of the individual.
To see the available commands, run:
facetool --help
The output will look something like:
usage: facetool [-h] {index,thumbnail,recognize_gui,recognize,detect,detect_camera_demo,track,download} ...
optional arguments:
-h, --help show this help message and exit
subcommands:
{index,thumbnail,recognize_gui,recognize,detect,detect_camera_demo,track,download}
index Indexes a dataset of face images.
thumbnail Generate thumbnails for an input image folder.
recognize_gui Detect and recognize faces in images or video.
recognize Detect and recognize faces in images or video.
detect Detect faces and write bounding boxes to stdout.
detect_camera_demo Detect faces on the webcam and show the results in a GUI window.
track Detect faces and tracks them.
download Downloads and extracts the LFW dataset.
To use the video face recognition software, you’ll first need to index a collection of faces to recognize. The software assumes that this collection is stored in a folder with one subfolder per individual, each named with the label you want to apply to that person. Inside these folders should be one or more example images of the person.
To index media run the following:
facetool index media/images media/index
to create a media/index folder containing the indexed files. Add a
--device=cpu to run this on the CPU. Indexing may take some time, especially
on a CPU (approx 30 mins).
facetool recognize media/example.mp4 media/index --device cpu (--device is an optional parameter)
The output will look something like:
6 [310 61 372 150] Aaron_Tippin 0.5715298652648926
8 [301 54 364 147] Bob_Geldof 0.526273787021637
10 [297 49 359 139] Jimmy_Kimmel 0.6068898439407349
12 [291 48 354 137] Jimmy_Kimmel 0.5837208032608032
14 [286 47 350 135] Jimmy_Kimmel 0.6029414534568787
16 [285 45 347 132] Jimmy_Kimmel 0.6005147099494934
18 [284 44 346 130] Jimmy_Kimmel 0.6477267742156982
20 [282 44 345 131] Jimmy_Kimmel 0.649223268032074
22 [279 43 343 133] Jimmy_Kimmel 0.6489613652229309
24 [277 45 341 132] Jimmy_Kimmel 0.6109910011291504
26 [277 44 339 130] Jimmy_Kimmel 0.6060200929641724
28 [274 44 337 129] Jimmy_Kimmel 0.5980086922645569
30 [273 45 336 129] Jimmy_Kimmel 0.6105645298957825
Each line has the format <frame-number> [<x1> <y1> <x2> <y2>] <label> <confidence>