Hi, Beta testers.
Today, we will release the 2.9 beta test app for both iOS and Android. You should receive this update within the next 24 hours due to the processing time with the app stores.
iOS version: 2.9.18
Android version: 2.9.20
RELEASE NOTES:
-
Added support for Cloud AI Person Detection (available only to a portion of testers but we will add more testers as we can)
-
Added support for hardware decoder for Android
-
Fixed auto-unlock accuracy bugs
-
Fixed a lock status accuracy bug on the Home tab
-
Other bug fixes
SUGGESTED TEST AREAS:
-
Hardware decoder (Android)
-
Cloud AI accuracy (for those already whitelisted)
-
Lock auto-unlock and status accuracy
KNOWN ISSUE:
- Zooming in/out of the live steam is not smooth on Android
Here is an FAQ about the cloud-based person detection we are now testing:
WHAT HAS BEEN CHANGED? HOW IS THIS DIFFERENT FROM THE PREVIOUS ON-EDGE SOLUTION?
XNOR’s solution is built for the CPU chip meaning that their person detection is run on edge. Our Wyze Cloud solution moves the detection operation from the Wyze Cam to our cloud (hosted in AWS).
DOES MY WYZE CAM SUPPORT CLOUD-BASED PERSON DETECTION?
With our new cloud-based solution, we were able to remove the limitation that prevented using person detection with Wyze Cam v1. The algorithm can be run on all Event videos sent to cloud from Wyze Cams v1, v2, and Pan.
CAN MY WYZE CAM SUPPORT CLOUD-BASED PERSON DETECTION AND COMPLETE MOTION CAPTURE AT THE SAME TIME?
At this stage, cloud-based person detection doesn’t support CMC videos due to some technical limitations. The Wyze AI team is actively working on the solution to this.
WHY DID WYZE CHOOSE TO USE THE CLOUD RATHER THAN USING ANOTHER ON-EDGE SOLUTION?
Compared with on-edge solutions, using the cloud means the computing resources are unlimited. This means that Wyze faces fewer limitations when designing AI features such as:
- Better model performance
- Running different models at the same time
- Supporting a wider range of devices and services
Considering those benefits to users, we’ve decided to go with a cloud solution first. But at the same time, we have also kicked off the effort of building an on-edge solution.
HOW ARE THE CLOUD MODELS BUILT?
The cloud person detection models were designed and built completely by Wyze engineers and scientists. Our data exclusively comes from the feedback videos submitted by our awesome Wyze customers for training the algorithm.
HOW DOES THE PERSON DETECTION BETA TESTING WORK?
The entire 2.9 test cycle will be a beta testing phase for cloud-based person detection. Wyze will gradually roll out the person detection to specific users. This is to ensure that our servers can handle the additional load. Please be patient with us!
But, if you cannot wait to try out the feature, we totally understand! Send us a request to join the waitlist and we’ll process it as soon as possible.
Link to the waitlist:
https://airtable.com/shrocE5N5h0CnXcoT
SHOULD WE STILL SUBMIT VIDEOS FOR TRAINING?
Yes, please do! To make our models more accurate, Wyze needs to feed the algorithm with many kinds of videos. From past experiences, we found that some scenarios have increased chances for false results. Those scenarios include:
- Indoor cameras with complex backgrounds: e.g. living room
- When an animal or pet enters the frame
So, in order to increase the diversity of the training dataset, you can help us by submitting different kinds of videos.
I TURNED ON MY PERSON DETECTION SWITCH, BUT I DIDN’T SEE THE PERSON ICON ON MY EVENT PAGE. WHAT SHOULD I DO?
Right now in the testing stage, there is an up to 1-hour delay after you change the status of the switch.
DID WYZE WORK WITH OTHER 3RD PARTY PARTNERS THIS TIME?
No, this time the solution is built in-house completely.
PRIVACY AND SECURITY CONCERNS
Q: Does the cloud-based person detection make copies of my videos?
A: No, that isn’t necessary for the person detection to function. After a video clip is uploaded to the cloud, the model will be run against the uploaded video to check for a person. The model reads the video frames and gives the detection results for each frame. During the process, there is no need to make a copy.
Q: Does cloud-based person detection save any personal information or features?
A: No, the cloud-based person detection only runs models to detect a person’s existence. It won’t save any information that can be used to identify a specific person.
Q: Who has access to the videos during the training process?
A: Voluntarily submitted videos can only be accessed by a team of Wyze employees who annotate the data. All of the employees have signed an NDA to keep the data secure. These videos are not viewed or shared with any other Wyze employees. Physically, Wyze also limits access to the data inside the Wyze office and it is only accessed by managed machines.
Q: What should I do if I see a large portion of false detection videos?
A: Wyze is continuously working to fine-tune the model. If many of your videos have incorrect results,
- try to adjust the angle of the camera to keep it as horizontal as possible
- try moving the camera so it is not looking through glass or a window screen
- check for an app or firmware update
And last but not least, send the video to us using the feedback button.
Q: What happened with the training data sent to XNOR?
A: XNOR destroyed the training data they previously had access to.
Q: Are the videos encrypted while stored by Wyze?
A: The videos are encrypted during transit.