Are you running the production app or a beta app?
Beta, but it was my understanding this was running separate from that.
We were instructed to use the production app for testing the AI.
Aside from that, beta apps normally do have bugs so , there’s a good chance the beta app has the issue.
I don’t have duplicate tags in the production app, and I also run Android.
I don’t currently have any beta apps running on any devices so I can’t check that.
It might be best to post the issue in the beta section
Thanks, I’ll run it up the chain that way. At least we’ve narrowed down where its happening.
There’s a post somewhere (I think facebook) where they said they had added the AI testing into the Beta app and it was ok to use that. I actually didnt sign up for this beta until that point because I didnt want to try to manage different software on different devices.
That might be the case, I didn’t see that
Is there a non-Facebook (I don’t have Facebook) explanation or instructions anywhere for how to use the beta testing tags to help you out with this? (I don’t mean how to enroll in it, I’m already enrolled.) For example, with this video:
It tagged it as a person, but there is no person, it is my black cat going in his kitty-door. So which of the following are the appropriate submission:
- Tag it as “pet” and submit it to tell it there was just a pet…and the AI will presumably figure out it shouldn’t have marked it as a person.
- Tag it as Missing Detection and Pet (to tell it that it missed marking it as a pet)
- Tag it as False Detection and Person (to say that it falsely identified it as a person)
- Tag it as both False detection (it said person when it shouldn’t) and Missing detection (it didn’t say pet, and it should have), and then mark both Person (the false detection) and Pet (the missing detection) and let the AI figure out which is which…or will that make the AI think it should’ve marked it as both Person and as Pet and screw it all up?
- Do I just choose one or the other, but not both (pick whether to mark it missing detection and pet OR false detection and person)…I can’t submit one then the other, it only allows 1 submission, but if I try to enter both the false and the missing at the same time, it could confuse the AI.
- Also, what to do about packages? Do I mark it as package, or package and person and face. What if the package was delivered an hour ago and a new event shows the package still on my porch, do I still mark it as package?
There are a lot of other potentially acceptable-seeming variations on this that would make sense, all depending on how Wyze intends to use the feedback…especially when there are multiple variations in play (a person delivers a package, may or may not see his face, and may or may not have my pet in the video, and may or may not have vehicles in the street in the background.
I am very happy to help with all this and submitting my videos to improve the AI, I would just like clarification on how I can best help Wyze, rather than randomly deciding on one of the above actions myself which might be the opposite of “helping” to improve things if it turns out the opposite was wanted and planned for.
I didn’t see any other thread on this topic, so it seemed most appropriate to ask here (it’s fine if Wyze feels it would be better in another section or it’s own beta thread or something)…It just seemed okay here since Wyze officially posted about it here in News and didn’t give any instructions on how they’d like us to help.
I look forward to an official answer and I will provide a lot of help with this to train your AI the way you want it trained.
I’m rather surprised that this clip was correctly flagged as a package given the low lighting and how brief the box was visible.
What the heck, that’s extremely impressive! I hope you submitted it with the “package tag” to validate it that it made the correct identification so it knows that was a job well done, rather than just a guess.
Mine is still messed up horribly and useless. Raccoons are setting off car detection (in an area that there aren’t cars), table lamps turning on inside are triggering vehicle, cat is either “me” or a person, Wall triggering as package. I wish they’d fix this thing but I do send in logs on all of it
I did end up tagging that as correct. I have only been running the AI for the last week and I would have thought it was just lucky except that I probaby have over 100 events of people walking that same path, mostly in full daylight without them being tagged as a package. I have had 4 package flags in that period (one was a a very obvious self placed one just to test) and they were all correct. I have had delivery people with packages walk that same line, never being in full view (obscured by the fence and plants) that did not flag as a package. I did not expect obscured views to correctly flag nor did I try to report those as a package since I didn’t think they would be good samples to train the AI.
My experience has be good so far. I have not noticed any glaringly bad incorrect detections yet. As I mentioned above I probably have over 100 events now. People in full view of the camera are consistently people. People walking on the other side of the plants and fence and are more than 50% obscured are sometimes not flagged as people. I had one event of a neighbors cat run through my yard and that correctly flagged as a pet. Vehicles I do not have much experience with as the camera does not overlook an area where vehicles go but I stuck a toy car in front of the the camera and it correctly flagged as a vehicle. Faces I have not thoroughly tested yet.
Do you have an example of a racccon triggering an event with a vehicle tag? My guess would be there is something in view that might resemble a vehicle when the racoon trips the camera.
Nice. There were a couple of frames where the classic “bent ams holding package” were visible but it was just a glimpse.
I’m pleasantly surprised. Realistically I would not expect consistency for a similar situation though.
My WCO and 1 other cam OFTEN detects my cats as a person. I can’t decide whether to mark it as “Missing detection” and “Pet” or as "False detection and Pet (falsely detected as person, and should be labeled as pet) or false detection and Person (falsely detected a person), etc.
I have started to get more appropriate detections in just the last 2 days. For the first few days I got 0 vehicle detections and 0 pet detections, and suddenly I get some showing up now without me doing anything different, so it is definitely learning quickly and well.
I have labeled the detected faces of myself, my wife and my kids, but I can’t do anything with those. There is no label of faces on the videos, no alerts, no sorting, nothing. It seems the facial detection does nothing at all right now. I can attach faces to my family and now it won’t show any “unrecognized faces” in the faces tab anymore…but it doesn’t seem to have any other purpose thus far. I’m sure they just haven’t programmed anything for it. If you or anyone else is able to do anything at all with the facial recognition, please let me know, because I’m just assuming there is nothing possible yet besides putting a faces in a saved profile that does nothing yet.
I have no pets so I’m pretty safe from false person alerts on that front. I’m not really sure the best way to tag those. Maybe just tag pet but not tag person since you can choose multiple tags?
I think I read somewhere the facial AI wasn’t working on the WCO yet (not totally sure about this) . My cam plus licence is on a V2 and it is picking up faces and tagging the videos with the named faces it recognizes. I can also filter by specific faces using the filter icon.
Oh! That’s where I have to go to set it. I tried to set it from the main event screen where it says “Faces” next to Person as a filter, and when you click there, it says “You haven’t marked any person yet” and the “Manage People” link there doesn’t work (have to manually go to Account tab to manage faces). I didn’t think to try the filter icon. It is there, and it actually works GREAT! I am really impressed. I can’t believe that it recognized our faces in some instances (I couldn’t really recognize the face, but it still labeled us correctly…I knew who it was because of clothes, time of day, and many characteristics I know. Crazy how good it is! Thanks for telling me I have to do it through the filter icon…it works great there…guess they need to fix the main event screen selector still.
I have Cam Plus on V2’s, V3’s and a WCO. You are right about the WCO not doing faces yet. It will do everything else, just not faces. I find this to be strange, since the video for it is processed on the same cloud with the same AI. I’d assume the AI would be told to check all videos the same way. The processing isn’t done locally on the camera, so I don’t understand any reason to exclude WCO videos. It makes no sense, but it is true, it doesn’t work. Not a big deal. I’m excited how well the Facial recognition sorts my videos so accurately! It’s impressive.
Now, they need to add 3 features to augment all this:
- Notifications for particular people (send me a certain kind of notification if my daughter enters my office, but not when I or my wife goes in my office).
- Allow reverse filtering. For example, show me all videos that don’t have ME in them. Or show me all videos that don’t have me, my wife or my kids in them (and notifications for any person detection except for my family).
- Person announcements. “Mom and Dad at the Front door” or something like that.
Those would be AMAZINGLY useful.
The main event page buttons are a bit wonky for me as well and on closer inspection it doesn’t appear like my pet, package or vehicle filter options work. Event tagging is working for all categories but it appears as though only person and face filtering are working properly.
As great as all this sounds I have to think the effort is misplaced - because getting any of these very cool notifications 15 seconds late is a very suboptimal experience. I love the person detection announcements but they would be so, so, so much more useful if they took 1 second instead of 15-20 (at best). Meaning… the V3 should have had local PD and their CamPlus efforts could be redirected.
That would be really good, and they’re even working on bringing the notifications down to as low as 4 seconds. But, I just don’t think 1 second is possible for AI things like a person, car, pet etc, because I’m pretty sure the AI has to quickly identify if the motion is either one of those and then once it verifies it, it then sends the notifications through the cloud, which is required for the notifications and current AI. Would be awesome if they figured out a way…
I thought the 4 seconds was going to be for motion only, not PD? I have another brand with on board PD but have not yet done a careful timed test to see what it can do.
I don’t remember if its motion only but I think it’s for both eventually. Maybe if a wyze employee sees this they can verify…