Action and script controlled automated training of neural network to readout the value of a digital counter. This is an via GitHub Actions automatized version of https://github.com/jomjol/neural-network-digital-counter-readout.
There you also find further background information.
- New training image of a digital counter is uploaded to the directory
/ziffer_raw/is uploaded - GitHub action for a training cycle including the new images is started
- starting point: old trained neural network (loading of the old training status)
- different versions of the neural network are trained in a dedicated docker container
- Standardized report are generated to evaluate the training success
- Reports and results are pushed to the GitHub for further investigations and usage
- Old training status is stored additionally (reports and neural network configurations)
The neural network is used within the "AI-on-the-edge" project to digitize different analog meters (watermeter, gasmeter, ...) as the a water meter measurement system. An overview can be found here: https://github.com/jomjol/AI-on-the-edge-device
- Accumulate new images
- New Images
- Techem images
- new images
- new images series (ISKRA meters)
- new images
- new images
- new images
- Corrected wrong labeling (LCD) and reset the learing
- Updated LCD images
- Full set of LCD images (rather pasty)
- Initial Version
Background and details for the neural network can be found in: https://github.com/jomjol/neural-network-digital-counter-readout