This project is based on a Masters Project that uses a json-defined model to generate a python CNN class.
The required modifications are, that the ipynb script is separated into train and sort.
The directory currently is:
.
├── categories
│ ├── cat1
│ └── cat2
├── ml-projekt-skater-erkennung.ipynb
├── model.jsonc
├── Requirements.md
├── requirements.txt
├── sort.py
├── suggestions
├── tosort
└── train.py
Where train takes images from categories/<category>/*.{any image file}.
Sort uses all images in ./tosort/ and places them into ./suggestions/<category>.
Note that we have to scale all images before using them to train or sort. A simple stretch-rescale should still be good enough. Of corse, do not modify the images itself!