Replies: 3 comments 1 reply
-
|
This does not sound normal, not at all! But you need to give us more information!
In the new release |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
Thanks for the response! Sorry that i did not add more detail, my intention was to bring in a possible idea not an error report. I agree too that this looks like a hardware issue on my side not a software problem. I've just updated to the fresh
My config is: [TakeImage]
;RawImagesLocation = /log/source
WaitBeforeTakingPicture = 2
;RawImagesRetention = 3
Demo = false
Brightness = 0
Contrast = 0
Saturation = 0
Sharpness = 0
LEDIntensity = 10
ImageQuality = 18
ImageSize = VGA
Zoom = false
ZoomMode = 0
ZoomOffsetX = 0
ZoomOffsetY = 0
Grayscale = false
Negative = false
Aec2 = false
AutoExposureLevel = 0
FixedExposure = true
[Alignment]
InitialRotate = 0.0
InitialMirror = false
SearchFieldX = 20
SearchFieldY = 20
AlignmentAlgo = default
FlipImageSize = false
/config/ref0.jpg 264 118
/config/ref1.jpg 205 268
[Digits]
Model = /config/dig-cont_0620_s3_q.tflite
CNNGoodThreshold = 0.5
;ROIImagesLocation = /log/digit
;ROIImagesRetention = 3
main.dig1 200 156 24 43 false
main.dig2 232 156 24 43 false
main.dig3 264 156 24 43 false
main.dig4 296 156 24 43 0
main.dig5 328 156 24 43 0
[Analog]
Model = /config/ana-cont_1207_s2_q.tflite
;ROIImagesLocation = /log/analog
;ROIImagesRetention = 3
main.ana1 398 220 75 75 false
main.ana2 362 308 75 75 false
main.ana3 273 344 75 75 false
main.ana4 178 305 75 75 false
[PostProcessing]
main.DecimalShift = 0
main.AnalogDigitalTransitionStart = 9.2
PreValueUse = true
PreValueAgeStartup = 720
main.AllowNegativeRates = false
main.MaxRateValue = 0.02
main.MaxRateType = RateChange
main.ExtendedResolution = true
main.IgnoreLeadingNaN = false
ErrorMessage = true
CheckDigitIncreaseConsistency = true
[MQTT]
Uri = mqtt://192.168.188.249:1883
...
;[InfluxDB]
;Uri = undefined
;Database = undefined
;user = undefined
;password = undefined
;main.Measurement = undefined
;main.Field =
;[InfluxDBv2]
;Uri = undefined
;Bucket = undefined
;Org = undefined
;Token = undefined
;main.Measurement = undefined
main.Field = undefined
;[GPIO]
;IO0 = input disabled 10 false false
;IO1 = input disabled 10 false false
;IO3 = input disabled 10 false false
;IO4 = built-in-led disabled 10 false false
;IO12 = input-pullup disabled 10 false false
;IO13 = input-pullup disabled 10 false false
LEDType = WS2812
LEDNumbers = 2
LEDColor = 150 150 150
[AutoTimer]
AutoStart = true
Interval = 1
[DataLogging]
DataLogActive = true
DataFilesRetention = 3
[Debug]
LogLevel = 3
LogfilesRetention = 3
[System]
TimeZone = CET-1CEST,M3.5.0,M10.5.0/3
TimeServer = 192.168.188.1
;Hostname = undefined
RSSIThreshold = -75
CPUFrequency = 240
SetupMode = false |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
Is it possible that your hardware setup is not fully encapsulated, meaning light from the outside has an impact? It looks like your last 2 images have an issue with the exposure. How did you fetch those images? Looking on the config, i don't see an issue. Only oddity is that you set the |
Beta Was this translation helpful? Give feedback.
-
|
I use https://www.thingiverse.com/thing:5167976 which makes it unlikely that external light source affects the exposure. Might be the images with wrong exposure happen while I also manually trigger some actions leading to this side effect? Usually I just use the Auto Timer with a Round Interval of 1 Minute. I've increased it a bit a while ago - but this did not help. The In the file server I recognize that the distorted images are smaller in size than correct ones > 30KB is OK, < 30KB are broken. Here a debug log and image of a distorted image with version: v15.7.0
Can it still help if I downgrade to |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I get frequent (let's say 1 out of 10) images that are obviously broken to the human eye. Images are cut in half, vertically shifted, or have severely wrong exposure. It seems like AI on the edge tries to complete each round whatever it takes, even with completely distorted images, and only aborts if the flow check gives a negative result. I'm not an expert in this, but I would expect that each step taken can indicate a false (or likely false) result. Starting from the image capturing process, through the alignment to the actual recognition. If the flow is aborted at the 1st stage that indicates bad quality, a lot of false readings could be avoided (in my case).
I don't know the details - so I might be completely wrong, but at least the alignment should be able to detect how well things went. A threshold to abort recognition if alignment was above XX° / pixel off might also be able to catch these cases.
Thoughts?
Edit: Formating
Beta Was this translation helpful? Give feedback.
All reactions