Quick project to train a TensorFlow Litemodel on a directory of classified images. Based on the official tensorflow notebook https://www.tensorflow.org/lite/models/modify/model_maker/image_classification
Can't get tf model maker lite to install on my dev box, so instead I've used this docker image and guidelines
the script 'main.py' will run on the docker to train a model from folders of images named after their classes. It should use your GPU if you have one, and your CPU if not. Should!
I've also included script 'image_capture.py' that will open up your default webcam and record images for each class - saving them in the appropriate format and location for you to later use in the trainer code.
For whatever reason I can't get the docker to read it's local file storage unless it's just downloaded the files itself. I therefor have used the same tgz zip/upload/download/extract workflow that's in the example code, and that seems to work fine - even if it does add a few minutes to each run.
Written with StackEdit.