Desperate - recognition doesn't work #3356
Answered
by
SybexX
dominikaherbst
asked this question in
General
Replies: 1 comment 4 replies
-
You could try version 16.0.0-RC4 and then use the zoom function to increase the ROI size. For me, the size 136x58 turned out to be the best. |
Beta Was this translation helpful? Give feedback.
4 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi,
I am close to giving up.
With some digits the recognition works, but then some digits change and suddenly a 4 is recognized as a 1, a 3 as a 1, a 2 as a 7 ...
I believe my problem might be that the digits do not nicely center on the meter. Some digits might come to rest in a slightly higher or lower position than their neighbors ...
I enabled and disabled "Synchronize y, Δx and Δy between ROIs", same with aspect ratio. I change the ROI positions and adjusted them as described. I even checked whether my numbers are already in the models (https://jomjol.github.io/neural-network-digital-counter-readout/) and they are (I think).
In my opinion the ROI boxes are too large, but to fit the inner box exactly to the number, they have to be that large.
I tried all models, but I don't get any model consistently working. Sometimes a model works with the ROIs, the reference image that I took, and numbers. But then I take a shower, some digits change, the recognition fails, and I am off by 500-1000L ... Sometimes that even happens with numbers that haven't changed.
It took me a couple of weeks to get everything installed without reflection, ordering extra parts, etc., and now I am close to giving up ...
Can anyone give me some advise?
Thanks!
Build Info
Firmware version: | Release: v15.7.0 (Commit: 0d0b018+)
Firmware build time: | 17.02.2024 00:15
Web interface version: | Release: v15.7.0 (Commit: 0d0b018+)
Beta Was this translation helpful? Give feedback.
All reactions