Hara Seiko
Aug 8, 2017 · 1 min read

How to create an image classifier that takes not big INPUT_SIZE like 64 or even 32

It would be great to crop objects without neural net but using cascade classifiers (LBP/Haar) trained for triangle/circle/rhombus objects (for example traffic/road signs) and then classify using TensorFlow trained model, 64x64 px should be enough because cropped images contain only our interested object