Alright that's in the latest image now and I modified the data preparation and verified it's correct in the cloud. It's gonna show the update message because the versions don't match, but you can just close the alert. I'll make the release soon.
See previous message. Just click out of it. Notice it says you have 1.8 and 1.7 is the latest. The version checking just checks inequality because you would normally never have something greater than the newest published.View attachment 216994
I updated again to get the confirmation feature, looks like that confused the update script.
Does this latest version show the version info somewhere? I'm not finding it
Yeah that will be totally different and it shouldn’t actually even try to create it if it doesn’t have the data it needs. I’d probably leave it turned off if you’re using platerecognizer, but even if it were on I don’t think it would do anything.I'm not presently using the ALPR module to do my plate reads; using the free PlateRecognizer instead (<2500 plates per month). I'm also feeding in my delivery vehicle detections as if they were plates.
What effect will that have on the data sent for training? My delivery data isn't truly a plate. And my plate data from PlateRecognizer doesn't have the position of the recognized plate (unless Ken changes something).
Also, what happens if I've corrected a plate that was only partly in the image. I know the rest of the digits from looking at the footage, or recognizing it as a known vehicle. But could that cause the model to hallucinate digits for you others that aren't there?
I just don't want to be messing up the training data.
Yeah that will be totally different and it shouldn’t actually even try to create it if it doesn’t have the data it needs. I’d probably leave it turned off if you’re using platerecognizer, but even if it were on I don’t think it would do anything.
Correct. I’d still leave it turned off because it’s just a different thing, but no there’s nothing it could do to mess anything up.So are you saying only data read using the ai_dump tag, will have the requisite info for training and I don't have to worry about my reads from affecting things if I still use the memo or plate calls?
Added manual confirmation also. Correcting a plate will also set confirmed to true.
The resulting training data will be organized into two categories: validated, unvalidated. That should be more accurate.
View attachment 216982View attachment 216983
Were you able to get it installed/working as a PWA on your home screen and not in the browser?
Added manual confirmation also. Correcting a plate will also set confirmed to true.
The resulting training data will be organized into two categories: validated, unvalidated. That should be more accurate.
View attachment 216982View attachment 216983
It worked the first time, but the next plate that gets corrected does not flip the confirm bit. I can flip it manually, tho.
I'm not sure how that should look, but yes, I have the icon on my desktop, and it opens correctly. However, I still see the top and bottom bars of the browser, making it look like a browser rather than a standalone app.
That said, I'm not using Safari, my default browser on my iPhone is Brave. Could that be the reason?
Awesome! Thank you Charlie!
Two questions:
1) When the training sets get uploaded? Is it like the the non-confirmed sets are uploaded immediately after injections and the confirmed ones when they get confirmed?
2) Correcting a plate doesn't set confirmation flag true for me. Is this only me?
That might be it. You can keep brave as your default browser, but I think you have to do it in safari when adding it to the home screen.
1) Checked every 500 plates. it is all uploaded. They're just separated into two folders.
2) See #794
1) I’m not sure what you mean by this. Are you saying it waits until there are 500 unuploaded recognitions before uploading them all at once?
2) The flag doesn’t light up for me, even after refreshing the page.