Heartbeat is a Face Recognition app, that you can upload Images to find more Images with the same face.
This Project can be used for evil shit, but the main Purpose was to show how easy and dangerous it is to build a mass surveillance service.
http://heartbeat.mowoe.com is fed with Images from various social Media Sites, e.g. Instagram. You can upload a picture of yourself or anyone else and Heartbeat will try to find images with the same person on it. (Wait up to 30s, the heroku dyno takes this long to start up.)
Heartbeat needs a MySQL(-compatible) Database to store its faces and images. You need to create a user and a Database, the necessary tables are created by the peewee ORM itself. If you want to create the Database yourself or already have a mysql/mariadb instance running you can use this code:
CREATE DATABASE heartbeat;
CREATE USER 'heartbeat'@'%' IDENTIFIED BY 'heartbeat';
GRANT ALL PRIVILEGES ON heartbeat.* TO 'heartbeat'@'%';
FLUSH PRIVILEGES;OR You can use this command to spawn a docker container in which the mariadb database will be created accordingly.
sudo docker run -d --name heartbeat_mariadb \
-p 0.0.0.0:3306:3306 \
-e MARIADB_RANDOM_ROOT_PASSWORD=yes \
-e MARIADB_DATABASE=heartbeat \
-e MARIADB_USER=heartbeat \
-e MARIADB_PASSWORD=heartbeat \
mariadb:latestFinally you can start the docker container:
sudo docker run --name heartbeat \
-p 127.0.0.1:5000:80 \
-e DB_HOST=172.17.0.1\
-e DB_PORT=3306 \
-e DB_PASSWORD=heartbeat \
-e DB_DATABASE=heartbeat \
-e DB_USER=heartbeat \
-e DB_TYPE=mysql \
-e OS_TYPE=local \
mowoe/heartbeat:latestVisit localhost:5000 to use the created heartbeat Instance.
By default heartbeat stores all uploaded images inside of the container as plain files. This however can pose some problems: The files will be deleted as soon as the docker container is deleted, which will render the whole heartbeat instance useless. To combat this, external storage is currently supported in the form of S3-compatible storage buckets (e.g. AWS, min.io).
When using any object storage, please be aware, that the bucket needs to be named heartbeat-images.
- To use Local space use the docker variable
-e OS_TYPE=local. - To use the AWS S3 Storage change it to
-e OS_TYPE=s3.- You also need to specify:
-e AWS_ACCESS_KEY=awskey-e AWS_SECRET_KEY=aws_key-e AWS_REGION=eu-central-1(or any other region)
- You also need to specify:
Heartbeat supports other Bucket Storage Systems too, this is why you need to specify -e endpoint_url=http://s3.eu-central-1.amazonaws.com or any other Endpoint to an AWS S3 Storage like interface (like min.io).
From now on you can use the heartbeat instance and upload images.
- The main page
/is for uploading pictures to get checked and find similar faces. - To upload new images to heartbeat (which will be processed and faces found on them will be saved) visit
/upload_new
To actually use heartbeat, each image has to be processed by face recognition software. To unload some work off of the webserver, this is done remotely (while this of course can run on the same machine). The easiest way to do this is again by spawning a docker container:
sudo docker run --name heartbeat_worker \
-d -e HB_HOST=172.17.0.1 \
-e HB_PORT=80 \
mowoe/heartbeat_worker:latest(adjust ip and port accordingly, container will fail if HB_HOST is not in the form of example.xyz or 1.2.3.4, as well as containing / or other unnecessary characters)
This will spawn two threads which will constantly look for unprocessed images and then process them. This is still wip and a bit hacky so dont expect good performance or stability. Due to some lil fuckups this doesnt really work as intended, when there are below 10 images saved in heartbeat. Also, please be aware that face recognition will only work as soon as more than 5 faces have been recognized and saved to heartbeat.
After all images have been processed, the value-vectors need to be combined in one model. This will be a representation of a 265-Dimension Space, which will serve as some kind of lookup-table for the identification of faces. As the creation of the model can take very long, it is not done automatically. This means a face recognition request to heartbeat will fail, if you didnt train a model yet. You can train a model by going to http://heartbeat-host/admin. This is not advised for huge amounts of images, as the nginx will timeout. If you have large amounts of data, please use a script, which does the training by itself and then uploads the model to your Data storage. This script is still WIP and not published, which means heartbeat is currently unable to handle very large data amounts, but in the near future this will be fixed.
Heartbeat needs a MySQL(-compatible) Database to store its faces and images. You need to create a user and a Database, the necessary tables are created by the peewee ORM itself. Startup a e.g. mariadb (compatible) server and execute the following sql code:
CREATE DATABASE heartbeat;
CREATE USER 'heartbeat'@'%' IDENTIFIED BY 'heartbeat';
GRANT ALL PRIVILEGES ON heartbeat.* TO 'heartbeat'@'%';
FLUSH PRIVILEGES;Clone the repository:
git clone https://github.com/mowoe/heartbeat.git
cd heartbeatFirst you need to install all necessary python packages, which can take some time:
pip install -r requirements.txtCreate a new file, name it heartbeat_conf.json and add the following contents:
{
"object_storage_type":"local",
"db_type":"mysql",
"db_auth":{
"host":"localhost",
"database":"heartbeat",
"user":"heartbeat",
"password":"heartbeat",
"port":3306
}
}gunicorn -w 1 heartbeat:appIf everything went well, you should be able to access heartbeat via localhost:8000 and you are done!
From now on you can use the heartbeat instance and upload images.
- The main page
/is for uploading pictures to get checked and find similar faces. - To upload new images to heartbeat (which will be processed and faces found on them will be saved) visit
/upload_new
To actually use heartbeat, each image has to be processed by face recognition software. While it is advised to do this work on a more powerful machine, you can use the supplied script to do this on you machine:
cd heartbeat
python face_encoding_example_worker.py localhost 8000 1(Last argument is the amount of threads to be spawned, on less powerful machines or when not working with huge amounts of images, this should be kept low)
After all images have been processed, the value-vectors need to be combined in one model. This will be a representation of a 265-Dimension Space, which will serve as some kind of lookup-table for the identification of faces. As the creation of the model can take very long, it is not done automatically. This means a face recognition request to heartbeat will fail, if you didnt train a model yet. You can train a model by going to http://heartbeat-host/admin. This is not advised for huge amounts of images, as the webserver will timeout. If you have large amounts of data, please use a script, which does the training by itself and then uploads the model to your Data storage. This script is still WIP and not published, which means heartbeat is currently unable to handle very large data amounts, but in the near future this will be fixed.
If you completed all of the above steps, you are now able to use your heartbeat instace and start uploading images to it! Congratulations!
Some Flowcharts to show you how heartbeat works internally.
Phase 3: The worker processes the image (face recognition) and submits the result (mathematical representation) back to the heartbeat server.
To upload an Image via the API, you have to supply the URL to the image, direct Upload is currently only via the Frontend supported. You also have to supply an origin of the image, so it can later be traced back. If you have any other information about the image, you can supply them via the "img_info" Key. This is just a JSON Object with all Infos about the Image, which can also be used later for tracing the Image back.
POST /api/add_image HTTP/1.1
Host: heartbeat.mowoe.com
Content-Type: application/json; charset=utf-8
{
"img_url": "https://example.com/example.png",
"img_info": "{'uploaded_date':128370}",
"origin": "example.com"
}To request Work form the Server you just have to supply a "work_type" Key, as Heartbeat theoretically also supports other recognition types than just Face Rec
GET /api/request_work?work_type=face_recognition HTTP/1.1
Host: heartbeat.mowoe.comTo submit the requested Work you have to supply the work and the image_id, that was retrieved when requesting work.
POST /api/submit_work HTTP/1.1
Host: heartbeat.mowoe.com
Content-Type: application/json; charset=utf-8
{
"result": "[representation of the face in a vector]",
"image_id": "12345678",
"work_type": "face_recognition"
}