the project uses Baidu Paddlepaddle platform to train pics and identify them using MobileNetV1 and other algorithms (see other branches).
- createdatelist
generate the list of dataset. In addtion, the pics should be stored in different folders according to their types before running it. - mobilenetv1
define the algorithm, which is mobilenet v1. - reader preprocess the images from dataset
- train
the program to build the model using CPU or GPU
you can use train-timestamp to pin each time for each 100 batch in default
- process
preprocess using gauss filters or converting to hsv space using process-hsv - infer
the file to identify the pic
you can use infer-multi to indentify a batch of new pics in the same folder
- gui
it can realize the identifacation using screenshot with a gui
as it says that, it is really primitive and written for just "it seems that we need a gui"