The code is for the competition RMB code recognition and my rank is 13th in the final
I used foveabox to detect the object box and crnn to recognize the RMB code,so I trained the detection model and the recognition model respectively. And cascade them in the test time.
-TMDetector #the part of detection(i.e. foveabox)
-TMBackbone #the backbone
-TMDetectConfigure #the configure file for train and test
-TMDetectDataSet #the dataset
-TMDetectLoss #the loss
-TMDetectUtils #generate train target for foveabox
-TMRPN #the rpn head
-TMTest #the test code of the detection part of the test
-TMTrain #the train code of the detection part of the train
#the part of crnn
-TMTextLine
-TMNonLocalNet #the NonLocal Modular
-TMTextLineConfigure#the configure file for train and test
-TMTextLineDataSet #the dataset for crnn
-TMTextLineNN #the model for crnn
-TMTextLineTest #the test code for crnn
-TMTextLineTrain #the train code for crnn
-TMEndToEndConfigure #the end-to-end configure file while includes detection and recognition
-TMEndToEndTest #the end-to-end test code which includes detection and recognition