Inspiration
According to a research conducted across 3 countries, at least 600 million people suffer from online image abuse, and one-third of them have considered suicide. Online image abuse kills people and it’s difficult to tackle. When it comes to getting consent, there is a disconnection between people and platforms. Companies like Facebook spent billions of dollars on content policy, but are still unable to determine whether an image is being shared consensually or not. When these platforms fail to act on abusive content, federal fines, and class action lawsuits happen. For victims, due to the disconnection from platforms and lack of access to proper tools, justice is rarely served.
In today’s online environment, everything is becoming more public, accessible, and transparent; inevitably, our privacy is jeopardized. Catfishing, identity theft, and cyberbullying are becoming more commonplace while technology and social media companies are not equipped to deal with the repercussions they might face, such as lawsuits and damage to their brands (Meta was fined $5 billion by the fed for privacy violations). I saw this directly in my previous role on the compliance team at Eventbrite, where the biggest issue we saw was IP related. As a platform, we weren’t able to determine if something is infringing or consensual unless we communicate directly with the owner of those pictures. The inability to track and determine consent is an issue we face across the industry, which is why online image abuse is so difficult to tackle.
To learn more about Alecto AI, please visit link.
What it does
Alecto AI is a facial recognition tool that identifies online image abuse (deep fake, fake profile, revenge porn, identity theft, IP infringement etc.). People can use Alecto AI to search the web for instances where their photos or videos are posted. This allows our customers to see if and where their likeness shows up on social media accounts, online dating profiles, or even pornographic/explicit content. Alecto AI serves as the broken link between individuals and content hosting platforms.
By allowing individuals to confirm ownership or consent of their accounts or materials, we are empowering them to retake control of their image online. For instances of online image abuse, we work with the tech companies (such as Meta, Twitter, Tinder) to act as a consent-based content moderation tool to ensure that our customers have a clear and easy path to remove unwanted uses of their images.
For the hackathon project, we developed a web-based reverse image search tool to showcase our capabilities in facial recognition and image search. We utilized Firebase storage to house our personal photos and selected celebrity face samples obtained from an open-source database. The facial recognition task was performed using the Facenet model from OpenCV. In particular, the Facenet model generated facial embeddings from a user-uploaded photo, which was then used to cross-verify against all facial embeddings stored in our MongoDB database. This process enabled us to determine if the uploaded face corresponded to any existing faces in our database.
How we built it
We utilized MongoDB as our primary database to maintain the names of all images and their corresponding facial embeddings, all of which are securely stored in our Google Firebase storage. We hosted our MongoDB database powered node js backend on Google Cloud Run so that we can access it through HTTP request. We also wrote our own Flask API that was hosted on Google Cloud Run to get access to OpenCV's Facenet model. The web app front-end is written in react and the back-end is written in node.js. We used Google Firebase Auth to perform authentication. Finally, we used Google Firebase Hosting to deploy and host our web app. Here is the link to the demo app: link
Challenges we ran into
In the beginning, we tried to import our face embedding dataset through JSON on the MongoDB GUI console, but it did not work the way we expected. So, we wrote a node.js program to format the JSON dataset and write it to the MongoDB database.
The IP setup and access in MongoDB are also making us struggle a little bit when we are trying to access MongoDB through Google Cloud Run instead of our local host. But we eventually figured out the problem through the online forum.
Accomplishments that we're proud of
Wrote and deployed a facial recognition reverse image web app with react, node.js, MongoDB and Google Cloud.
What we learned
We learned MongoDB as a database solution. We also found that Google Cloud Run is a great service that we can deploy and access customed API easily.
What's next for Alecto AI
User Acquisition Work with social media companies such as Facebook, LinkedIn, Twitter; online dating companies such as Tinder, Hinge, OK Cupid etc. to add consent & identity features for end users
Log in or sign up for Devpost to join the conversation.