AI Detection Errors

I’m sure this has been discussed, but I cannot find the proper thread. The AI detection has serious issues. First of all, if I have something selected (like vehicle) if there is any movement, and there are vehicles in the frame, it will say a vehice is detected. So, I turned that off and have people and pets selected, but I am getting notifications that moving vehicles are people, and there are no discernible people in the frame.

1 Like

Currently the ai scans anything in the detection zone, whether it’s moving or not. Wyze has been working a lot on changing this, but for now that’s a limitation.

Send some screenshots of when it detects a car as a person, as there could be a diffrent object in frame it’s seeing.

Ai is not perfect and never will be, but please submit these false detections and tag them appropriately with the little banner at the bottom of the event video page.

1 Like

Thanks @IEatBeans and @Rulwiz for the feedback. I have been and continue to tag the videos.

A couple of questions…is the AI learning specific to each camera? So for instance, if I submit these tags, does it help that individual camera learn overtime as well as help others?

1 Like

As far as I know the videos are submitted into a large pool, and monthly (unless needed more often) they retrain the ai on the new larger pool.

If you submit enough videos to make up a decent chunk of the pool, the ai will use those videos (correctly marked).

so I’m pretty sure it’s the same for everyone, and for rulwiz not sure why it updated so quickly, he must have just gotten lucky and was near the regular update or if they did an extra one as they do sometimes.

Alright. Thanks! For clarification, when I am providing feedback, am I selecting whatever is moving in the video or whatever is in the view of the camera? For instance, someone is walking by my camera and there are stationary vehicles in the frame, so it tags it as vehicle and person…should I submit just the person tag?

I think you should submit everything. Wyze is working on moving object detection, so things may change, but for now you should tag everything in the video.

We had an incident at our next door neighbor ‘s yesterday and multiple responders, rescue vehicles,ambulances, went down our driveway to help. There were maybe 15 vehicles involved and they all went past 2 of our garage mounted cams within 25 feet. NOT A SINGLE VEHICLE WAS TAGGED. It did pick up some people and a few car door closing sounds. I’ve been sending in AI corrections daily for at least a year and it still won’t recognize any vehicles slowly driving through.

Can you send the video it did capture so we can see what the area looks like.

Also, what cam is this?

Do you have CP on this cam, and vehicle detection enabled?

Did it capture a video just didn’t tag it?

The video is too large to put here. How would you like me to send it?
Cam+ is used on both V3 cams and all the events are selected on both including vehicles.
It captured all the videos but none were tagged as vehicles.

Can you upload it to YouTube unlisted and send the link

Do I need a YouTube account? If so, I do not have one.
I’m learning how to get the videos on my iPad to come out shorter. Let me see what I can accomplish.

Here is a short one showing 4 of the vehicles that arrived (mine are the 2 stationary ones at the bottom).

Vehicles leaving:

One more from the other cam.

2 more headed out:

Can you verify you don’t have any detection zones setup? Also, send a screenshot of your cams event recording settings, and an additional screenshot of the “Smart Detection” settings page. Thanks!

Interesting, I turned off and turned back on vehicle detection on both cams and I now have a ton of vehicle videos. Looks like it tagged my stationary cars every chance it could. I have turned them off, as this is now useless and annoying. I will have to muddle through them all to see if the few vehicles that drove past were correctly tagged.

I have never used used detection zones but it looks like that will be the remedy to block out my parked cars.

I monitor one of the cams 24/7 to see who may be up by our garage and I notice that one cam’s audio will not stay on. I keep having to turn it back on quite a bit lately. Sometimes my audio setting will effect others in the group. The other cam in that group has very bad audio so whenever I try to listen to it I have to turn it off.

Looks like I have some work to do…
Thanks - Paul

Maybe the AI needs more ambulance/firetruck/vehicles with warning lights examples for training? That was my first thought about the uniqueness of those, Maybe even the busyness of the livery on the side of the trucks played a part of the missed tags, i dont know just saying what my first thought was.

The way that the AI object tagging works is that the camera detects motion, and when the motion is not excluded by a detection zone and is exceeding the threshold of the sensitivity setting you set, it uploads a clip to the cloud. Once in the cloud, the AI analyses the whole video for recognized objects. The AI tags aren’t necessarily the detected motion, the AI tags are what’s recognized in the clip. This is wither the recognized object is moving or stationary.

Also, the “detection zone” is only with in camera, or client side only, it does not get sent along to the cloud when the clip is uploaded. So the AI tagging of vehicles, even if stationary is the current function and is working as currently intended.

The current AI is already being worked on and upgraded, but voting on related wishlist items help support those and show the Devs/Engs that there is a folliwng that wants these things implemented. Like this one:

2 Likes

Yea, a detection zone will help remove those parked car detections, and then see if the moving cars get tagged reliably.

1 Like

I have always only submitted moving vehicles and untagged all my stationary ones. There are just too many to correct from yesterday after I switched the vehicle detection off then on.

Thanks for submitting all the videos! Just saying that currently the tagging of vehicles, wither moving or stationary is the way the current AI works. The motion that’s detected could be anything, moving leaves, spiders, etc, and if a vehicle is in frame it should be tagged by the AI.

1 Like

So far so good. Now I just need to find out how little of a zone I can get away with so it won’t see the cars.