Come join us in shaping the future of Wyze AI

Greetings, fantastic Wyze family! How’s it going? I’m Shawn, one of the Wyze product managers.

It’s been a month since we launched our first AI product, person detection, to our cams. In the past month, we received lots of constructive feedback and opinions, which are very insightful and helpful for a new player in the AI field for tapping into the applications of technology and understanding the scenarios. Without you all, we could not launch the first version. And without you all, we cannot figure out the direction of the next step. We believe that’s exactly how Wyze should shape its products. I’d like to take this chance to have a quick review of what we’ve learned from the past month. And, more importantly, we’d like to invite you to help the future of Wyze AI.

We released person detection ver 1.0 on July 9th and we are glad to see that this function helps filter out unnecessary notifications and event recordings. More than half of active users and devices are using the service right now. We also saw a great improvement in notification engagement. Seeing our products becoming #TGTBT (Too Good To Be True) is incredibly encouraging for us moving forward.

At the same time, unlike our other software products, AI cannot be built from scratch. Wyze and XNOR teamed up and started with the open images from the Internet at first. But there is no way to improve the accuracy from 50% to nearly 90% without real user data. The AI squad is so grateful to the community for their help with beta testing. We received tons of videos submitted voluntarily by you all. Even after the launch, we have been thrilled to continuously receive videos for training. Kudos to the community!

Despite achieving the first milestone and the successes in the past month, it’s even more important to us to keep our minds open to improvements and solving problems. Tickets have been coming in talking about how person detection failed to meet their expectations. This isn’t comfortable but it is an opportunity for us to grow. There are several great lessons we’ve learned from you. First, for us, 90% accuracy is different from other AI products in the market. Because in most cases the cam won’t be moved from place to place, the content it sees also won’t change. That means if the model fails to detect something for a single user, it doesn’t mean that it works 90% of the time. It’s a binary of “work” or “does not work.” It’s interesting to know cameras see your dog, vehicles, or even moths as people.

The second biggest criticism is that we should remove the 5-minute cooldown and trigger events only when a person is detected because you don’t want to miss a single person event. We’re glad to see that person detection makes Wyze Cam more useful in security scenarios (though it is not intended as a security camera). Admittedly, this task is more challenging than it looks since we have to deal with challenging logic at the framework level. But we’re excited to be pushed to extend the boundaries of our technology. Our dev team is working on all those problems intensively right now. Want to know the latest status? I’ll show you later on.

In addition to all the great feedback, the Wyze AI team is also looking forward to the future. There are lots of areas that AI can be applied. In the cutting edge AI industry, researchers are training machines to understand what they see (computer vision) and hear, respond back (natural language processing), guess what someone is thinking (prediction and regression), and even mimic human behaviors (automation). Bringing all this technology back to Wyze, we are seeing opportunities to have Wyze Cams recognize more objects (pets, vehicles, packages, birds, etc.), understand different sounds, find the events you interested in, pop out notifications at the right time, and so on and so forth.

It’s not limited to cameras. We want you to make full use of both Wyze hardware and software products more easily. As a result, we are exploring how AI can help setup devices properly, solve connectivity issues, and even perform some actions automatically. However, the most important thing to do first is to invite you all to the process. Which cases are most valuable? Which task has a higher priority? What solutions are expected from Wyze AI? You’re the one with answers to these questions.

Why are you still waiting? Come join us in previewing the upcoming solution, getting involved in the product development and strategy, and shaping the future of smart home technology! We’re always ready to listen. Please tell us what you want to see from Wyze AI in 3 months, 6 months, 1 year, or even longer!

Last but not least, the journey should be fun. Take a quiz about AI here and see how you score against your friends!

Cheers,

Shawn

2 Likes

:muscle:

2 Likes

So what would you like to see in the future, AI Supreme? :smiley:

1 Like

short term: other object detection, specifically animals and packages

long term: only record cloud clips based on AI detection settings

5 Likes

People tracking instead of motion tracking. When motion tracking is clicked it can pan which is outside will not move of trees and bushes to track people.

1 Like

Differentiation between people and my dog. That is my short-term dream… :relaxed:

5 Likes

I’d like to see the motion sensor have a setting for small animals that wouldn’t trigger an event.

1 Like

I would like to see the AI pick the best thumbnail view in the rich notifications. Right now, the thumbnails are usually meaningless.

5 Likes

Telling from the time, I think you are the 1st to win 100 score. Congrat!

1 Like

just be curious, for animals, what do you want to do? Record video only when detect a deer? Or notify you when your puppy run out of the room?

Remove or reduce the cloud cooldown, I would be happy to pay a small subscription for that.

To compliment that, allow the user to chose not to send “generic motion” to the cloud, and to only send to the cloud on person detection. Thus reducing how often they are in the cooldown. This should also help reduce traffic to Wyze servers too, videos of a leaf blowing by or a fly that rarely get watched. Traffic costs for Wyze, catching more of what we care about for customers. Win win.

More complex motion detection zones would be next on my list.

I would also love to see some sort of combined timeline for an entire group. Including sense sensors and even SD card recordings in that group. So I open the group and have a timeline a bit like the SD card playback view now, but with dots on that timeline showing me each cloud event over the top of what I have in the SD cards, each door open close, each motion trigger. So I can better piece it all together better when reviewing.

Finally a more basic version of the above on a per camera basis. So in the SD card playback view, show me in the chunks when a person was detected, or if I have it in continuous record, show me when there was actual motion. Different colours in the 1 minute chunks would work. Then overlay a button for the cloud clips too. All in the same view.

Great work, keep it up.

7 Likes

Guess what we are doing right now?

3 Likes

its possible to get a 110??

I’d like to see vehicle detection. The person driving rarely gets out in 12 seconds, so a vehicle is a leading indicator of a person about to drop a package or knock on your door (note my front door camera’s detection zone is halfway up my driveway, so I have no interest in the street, which is more than 20 feet away anyways).

6 Likes

bingo! you get it.

1 Like

Just got 120 on ai quiz

Mine says a 130 so I do not know what the maximum is, I still have it on the screen just not sure how to post it here

I would like to see longer cloud recordings when a person is detected. Also an alert when an SD card is not detected.

1 Like

Separate people detection zones and non-people detection zones.
For example, for the camera shooting down my driveway, I’d like to be notified a person anywhere in the frame, but limit notification of other objects to things that are near my door.

1 Like

I would like the contact and motion sensor events to be usable via Alexa. I use IFTTT now for that but that platform is terribly laggy and very buggy.

1 Like