And I don’t even mind it because I think it’s really kind of cute and funny.
I believe the bird is triggering a recorded motion event. Wyze AI probably thinks the vertical objects in view is a Person.
I ocassionally get a false Person notification on my camera below. I think AI thinks my mailbox or street light is a Person.
I have shared many videos with Wyze with the correct tagging for this camera. This seems to have cut down on the false Petson notifications for me.
1 Like
I get Possum ”Person” frequently even when the possum is directly in front of the camera. 
1 Like
I’d just name my mailbox. 
.
Thank you!
1 Like
I think that’s right. I don’t know which object is being seen as “Person”, but @theprojectfox could do some tests and figure this out using a process of elimination by creating non-person movement (waving a broom or something in front of the camera, or I’ve even seen suggestions of shining a laser pointer at a camera to trigger it for testing) and then eliminating specific objects by enabling Detection Zone and altering the excluded areas between tests. That might not ever result in an answer, though, because the AI might still evaluate the entire frame (including the excluded area) whenever a motion event is triggered. (I’m really not sure where Wyze is with that these days.)
Submitting these events back to Wyze to train the AI could be helpful, too.
2 Likes
While I admit that this is not particularly a scientific method. And that correlation is not causation. I will say that, anecdotally, it’s only ever mourning doves. . And to be fair I can see it. I’ve named mine. Bob. All of them.
It didn’t occur to me that it would be valuable for wyze to.get this reported every time. I will start being more diligent about that. Thank you again.
1 Like
I don’t know what the actual value is, but ostensibly those submissions from users help to improve the AI over time.
A tailless squirrel used to visit my parents’ backyard frequently. My mom named it “Bob”.
1 Like
From what I understand they use them to train AI. Not sure how effective the training is as it works for sometimes than after that all Hell brakes loose 
1 Like
This is also my understanding, but I don’t have insight into the process. I’ve seen @carverofchoice mention a…
…but I don’t recall seeing any official mention of a schedule, so I don’t really know how quickly user submissions are integrated into the model. I’ve had the perception at times that a batch of problem detections will crop up suddenly—based on what I see in the app and in what users report here in the Forum—and then sometimes the number of those reports subsides within a few days, so I don’t know what the true update frequency is.
2 Likes
I guess will never know, Wyze’s trade secret 
1 Like
Yeah, I’m not sure there is a set schedule for updates.
Another monthly update they do is the app. It has a monthly public update, but it’s not reliably scheduled for a certain day of the month. If you check the release notes, it’s all over the place.
I believe the AI updates are the same way. We only know that employees have told us that they do a monthly update for the app and a monthly update for the AI. A few times we have had an employee post to tell us the exact day a monthly AI update was being pushed out, but usually they are “silent” releases and we only know they happened because of the instant differences… Sometimes good or bad. It’s using machine learning, so Some changes can significantly impact detections within a single month, some are gradual and take a while to make a noticeable impact. In some cases, a single user submission may seem to have zero impact, while a bunch of similar submissions from different cameras at different times and different environments. Submitting the same kind of thing will be more impactful. But the training data already has hundreds of thousands or millions of examples, so new training data may take a little bit of time to influence the strong existing data. There are a lot of variables involved in machine learning.
But I have definitely seen submissions make changes. I have gone from having my black cat get reported as a person at night nearly every single time, to less than 1% of the time. It slowly went over the course of several months of updates from nearly 100% to less than 1%.
2 Likes