This is Shawn from Wyze! The Wyze AI team has been receiving more and more questions about what goes on behind the scenes to build our AI technologies. We want to keep you updated on what our team is up to and how it will impact your AI experience! So these messages are about to become a regular thing.
Actually, we’ve been testing this out for the last 2 months as we worked to make the info more accessible (tech jargon is fun but not the best way to communicate with everyone all the time). In case you are interested in reading the previous posts and discussions, you can find them here: March 2021 and April 2021.
This month, the Wyze AI model update mainly focused on improving Pet Detection. Pet Detection is ready to be launched after 6 months of pilot testing! We ran the final rounds of training for Pet Detection to improve the performance of the model last month. Specifically:
- Average precision (AP) for large objects (indoor pet monitoring use cases) reached our production level.
- For medium and small object detection, we’ve improved the AP by 66% (medium size) and 100% (small size) compared to what it was before. Though it’s still challenging to detect smaller pets or pets that are farther away (like pets on a walkway).
- Pet detection in night vision mode also improved.
You can play with Pet Detection by going to Account > Services > Cam Plus under the Active Subscription section. Select the device and turn on the feature. If you see any incorrect Pet Detection results, we’d really appreciate it if you submitted the video to us using the banner at the bottom of your Event video so we can use it for improving the models!
In addition to Pet Detection, we also added more feedback data to improve Person, Vehicle, and Package Detection to cover more edge cases.
If you enrolled in the Wyze Face Recognition test before, you may be aware that Face Recognition was shut down and undergoing a system migration from v1 to v2. This month, I’m glad to announce that after a full month of intensive migration work and internal testing, the Wyze Face Recognition service is opening up again for testers. The new system:
- Groups similar faces together so that you can quickly identify unknown visitors and watch the videos.
- Improves the accuracy of detecting qualified faces for recognition. Check out the FAQ here for guidance on improving detection.
If you were a Face Recognition v1 tester, the feature is ready to be used! Just go to the Cam Plus settings page and turn on the face recognition feature again. If you were not previously a tester and are interested in testing the feature now, you can sign up here.
In addition to the model performance, we are also about to test a new mechanism to improve the AI performance on Wyze Cam v3. If you are interested in testing, sign up here. More details will be sent through email if you’ve been added to the test crew and we’re ready to start.
To help us with training the new models and building new AI features, we are looking for the example data listed below. If you can find a few videos matching these descriptions in your list and are willing to send them to us using the video feedback tool, it would be super helpful for accelerating the improvement speed!
- Videos with babies inside the frame
- Videos with audio of a dog barking
- Videos with audio of a baby crying
- Videos with packages inside the frame
- Videos with animals inside the frame including but not only limited to dogs, cats, birds, and bears (we hope you don’t have a bear in your home but we have actually received videos of this before).
You can submit the video using the banner at the bottom of your Event video.
Like I said earlier, we’ve been posting these updates for the last couple of months. If you would like to check out the previous posts or see other Wyze News, please check out our forum category here.