da03 / Attention-OCR

Visual Attention based OCR
MIT License
1.12k stars 363 forks source link

Hello, I found that the same pair of images twice predicted results may be different, the reason for this problem is what? What is the current accuracy of the method you are tuned to? #44

Open zj463261929 opened 7 years ago

zj463261929 commented 7 years ago

The problem is solved. The problem appears ”cnn_model = CNN(self.img_data, not self.forward_only) #(True))”. svt accuracy:from 67% to 75.6%

zoe218 commented 6 years ago

@You mean that change 'cnn_model = CNN(self.img_data, True) #(not self.forward_only))' to ”cnn_model = CNN(self.img_data, not self.forward_only) #(True))”?