search

jomjol / AI-on-the-edge-device

Easy to use device for connecting "old" measuring units (water, power, gas, ...) to the digital world
https://jomjol.github.io/AI-on-the-edge-device-docs/
5.26k stars 580 forks source link

Digits interpolating direction #3051

Closed mkmer closed 2 months ago

mkmer commented 2 months ago

The Problem

It appears the system is interpolating "up" while my digits move "down". For example - my window shows 3/4 (half way between 3 and 4), the recognized digit is 4.5 (should be ~3.5)

Version

Release: v15.7.0 (Commit: 0d0b018+)

Logfile

[0d00h00m00s] 2024-04-27T19:06:42 <INF> [MAIN] =================================================
[0d00h00m00s] 2024-04-27T19:06:42 <INF> [MAIN] ==================== Start ======================
[0d00h00m00s] 2024-04-27T19:06:42 <INF> [MAIN] =================================================
[0d00h00m00s] 2024-04-27T19:06:42 <INF> [MAIN] PSRAM size: 8388608 byte (8MB / 64MBit)
[0d00h00m00s] 2024-04-27T19:06:42 <INF> [MAIN] Total heap: 4380199 byte
[0d00h00m04s] 2024-04-27T19:06:47 <INF> [MAIN] Camera info: PID: 0x26, VER: 0x42, MIDL: 0x7f, MIDH: 0xa2
[0d00h00m04s] 2024-04-27T19:06:47 <INF> [SDCARD] Basic R/W check started...
[0d00h00m04s] 2024-04-27T19:06:47 <INF> [SDCARD] Basic R/W check successful
[0d00h00m04s] 2024-04-27T19:06:47 <INF> [SNTP] TimeServer not defined, using default: pool.ntp.org
[0d00h00m04s] 2024-04-27T19:06:47 <INF> [SNTP] Configuring NTP Client...
[0d00h00m04s] 2024-04-27T15:06:47 <INF> [SNTP] Time zone set to EST5EDT,M3.2.0,M11.1.0
[0d00h00m04s] 2024-04-27T15:06:47 <INF> [SNTP] time zone: -0400 Delta to UTC: 14400 seconds
[0d00h00m04s] 2024-04-27T15:06:47 <INF> [SNTP] Time is already set: 2024-04-27 15:06:47
[0d00h00m04s] 2024-04-27T15:06:47 <INF> [MAIN] CPU frequency: 240 MHz
[0d00h00m04s] 2024-04-27T15:06:47 <INF> [SDCARD] Folder/file presence check started...
[0d00h00m04s] 2024-04-27T15:06:47 <INF> [SDCARD] Folder/file presence check successful
[0d00h00m04s] 2024-04-27T15:06:47 <INF> [MAIN] Tag: 'v15.7.0', Release: v15.7.0 (Commit: 0d0b018+), Date/Time: 2024-02-17 00:15, Web UI: Release: v15.7.0 (Commit: 0d0b018+)
[0d00h00m05s] 2024-04-27T15:06:47 <INF> [MAIN] Reset reason: Via esp_restart
[0d00h00m05s] 2024-04-27T15:06:47 <INF> [WLANINI] SSID: kasper
[0d00h00m05s] 2024-04-27T15:06:47 <INF> [WLANINI] Password: XXXXXXXX
[0d00h00m05s] 2024-04-27T15:06:47 <INF> [WLANINI] RSSIThreshold: -75
[0d00h00m05s] 2024-04-27T15:06:47 <INF> [MAIN] WLAN config loaded, init WIFI...
[0d00h00m05s] 2024-04-27T15:06:47 <INF> [WIFI] Automatic interface config --> Use DHCP service
[0d00h00m05s] 2024-04-27T15:06:47 <INF> [WIFI] Set hostname to: watermeter
[0d00h00m05s] 2024-04-27T15:06:47 <INF> [WIFI] Init successful
[0d00h00m07s] 2024-04-27T15:06:50 <INF> [WIFI] Connected to: kasper, RSSI: -69
[0d00h00m08s] 2024-04-27T15:06:51 <INF> [WIFI] Assigned IP: 172.16.2.55
[0d00h00m09s] 2024-04-27T15:06:52 <INF> [SNTP] Time is synced with NTP Server pool.ntp.org: 2024-04-27 15:06:52
[0d00h00m11s] 2024-04-27T15:06:53 <INF> [MAIN] Device info: CPU cores: 2, Chip revision: 301
[0d00h00m11s] 2024-04-27T15:06:53 <INF> [MAIN] SD card info: Name: SU04G, Capacity: 3781MB, Free: 3759MB
[0d00h00m13s] 2024-04-27T15:06:55 <INF> [MAIN] Initialization completed successfully
[0d00h00m15s] 2024-04-27T15:06:58 <INF> [LOGFILE] Set log level to ERROR
[0d00h02m32s] 2024-04-27T15:09:15 <ERR> [POSTPROC] main: Raw: 0005109.6, Value: , Status: Neg. Rate - Read: - Raw: 0005109.6 - Pre: 4005076.9
[0d00h16m00s] 2024-04-27T15:22:43 <ERR> [POSTPROC] main: Raw: 0005108.4, Value: , Status: Neg. Rate - Read: - Raw: 0005108.4 - Pre: 4005105.2
[0d00h18m39s] 2024-04-27T15:25:22 <ERR> [POSTPROC] main: Raw: 8005108.7, Value: , Status: Rate too high - Read: 8005108.7 - Pre: 4005105.2 - Rate: 1333334.5

Expected Behavior

I think it should detect 3.5 instead of 4.5. In attached shot, the number is 3 / 4 should be ~3.8 rather than 4.2 - because the number moves "DOWN" not "UP" in the snip.

Screenshots

image main_ana1

Additional Context

No response

mkmer commented 2 months ago

I guess this is not a "bug". Rather a need to train it with a different dataset :(