Most ways to implement faster rcnn are too complex, tens of thousands of lines of code dazzles us. So I tryed to implement faster rcnn by brief code. And meanwhile, making a more brief file architecture. Therefore, this code is really kind to the beginners.
ps: I refer to the smallcorgi/Faster-RCNN_TF. And I only use python which means the roi-pooling part is replaced by the function of tensorflow called tf.image.crop_and_resize.
So let's check the files include:
class_name.py # Include the class name list both in English and Chinese
model.py # The main file, include faster rcnn
train_model.py # About train this model
test_model.py # About test this model
mAP.py # About compute mAP for one trained model