self learning classifier

No need to train separately, learn the object features directly on the development board, and then use it directly

Demo video: youtube or bilibili


If use the lite version kmodel, you should add fea_len arg as 512 when create classifier object, when use the bigger kmodel this param is not needed:

classifier = kpu.classifier(model, class_num, sample_num, fea_len=512)

Then start learning objects after running

  • Press the boot button on the development board to capture 3 categories mobile, car, keyboard, each category only needs to be captured once
  • Then capture 15 pictures, no order is required, such as capturing 5 pictures of mobile phone, 5 cars, 5 pictures of keyboard
  • Then it will automatically learn the features of these 15 pictures
  • The last recognized image category will be displayed in the upper left corner

Save/load learned features

  • Use to save the learned features to the path file
  • Use KPU.classifier.load() to load features, refer to file