CompreFace 0.4: New Features, Better User Experience, and Privacy Protection
It’s been two weeks since we publicly announced the release of CompreFace, the face recognition service developed by Exadel. We’re now ready to announce the new 0.4 version of CompreFace, which includes major improvements to the face classifier method, as well as a new face verification feature and some new features for existing and upcoming users.
Software Development Times recently recognized CompreFace as the open-source project of the week! SD Times highlighted the solution’s key features, the way CompreFace executes face recognition, and the value it adds to businesses. You can check out their full blog post here to read their analysis of CompreFace or keep reading to see our explanation of the changes we’ve made and the features we’ve added for our newest release.
We’ve changed to a face classifier method that keeps the similarity score permanent
Known as the euclidean distance, our new face classifier keeps the similarity score stable no matter how many photos you upload to the face collection — ten, a thousand, ten thousand, or anywhere in between. With this change, it’s easier to choose the threshold of face recognition; it won’t get lower when you upload more faces to the collection.
We’ve also tested the accuracy of the face recognition neural network model that we use. It is 99.65% accurate according to the LFW benchmark test, which is recognized by developers as one of the top metrics in face recognition.
We’ve added a face verification feature that determines if a person is who they claim to be
In the previous release, we only had a face identification feature, which indicated whether or not a face already existed in the face collection. With our new and improved face verification, it’s easy to confirm that a person’s physical face matches the face ID uploaded to the face collection and that a person is who they say they are.
For now, we use the same face collection for face verification purposes. Check out the handy REST API that we created and referenced on our readme page.
We’ve added a Demo Page to help you get started with CompreFace more quickly
We understand that new users want to know how CompreFace works before creating an account and reading through the documentation. That’s why we created a simple Demo Page, where you can recognize a range of celebs faces to see how the recognition process works. To do so, you just need to install CompreFace in your system and click “Try Demo” on the start screen.
We’ve added a test page where you can try facial recognition with your own images
The test page is helpful when you want to test how the recognition process works with faces from your face collection. For example, you can upload a photo of someone wearing a mask or someone whose face is partially covered. After you can see if CompreFace will recognize it, or you can test how well CompreFace recognizes your face among the other 10,000 faces uploaded to the face collection. You can also discover how REST request and response appears at the endpoint.
We’ve added a new feature that disables the storing of photos in the CompreFace database
We take data protection very seriously and we want to comply with privacy laws like the EU’s General Data Protection Regulation and the California Consumer Privacy Act. That’s why we added a new feature that allows users to disable the setting that saves photos to the database.
When users choose to disable saving photos, we only keep a limited amount of information in the database:
- Face Embedding: 512 numbers calculated by the neural network which are used to classify face features. It’s impossible to restore original faces with those numbers.
- Subject: an ID that you create when adding a face to the face collection. It can be either a person’s name or random symbols to keep anonymity.
You can find more information about our latest release of CompreFace on our GitHub page.
Originally published at https://exadel.com on December 17, 2020.