what is this global variable 'NUM_CLASSES' in lstm_ocr_ctc?

[wangx404]

In lstm_ocr_ctc.ipynb, this is a global variable NUM_CLASSES, but it was not used. So is there something wrong here?

[wangx404]

I find another probelem. In line_segmentation.ipynb, after defining the SSD network, you invoke in like SSD(2). However, when we dive into the network structure, we can find this statement gluon.nn.Conv2D(num_anchors_predicted*(self.num_classes + 1). According to original define, this is wrong! Either you add another one class to calculate num_classes, or you add one in defining the cnn network.

[jonomon]

Hi @wangx404,

Thank you for pointing this out.

For the first point, lstm_ocr_ctc.ipynb is actually outdated and will be removed in future versions. Please refer to handwriting_ocr.ipynb for a more up to date version of handwriting recognition.

For the second point, thank you so much. We will look into this!

[ThomasDelteil]

@wangx404 I deleted the old outdated file for clarity, please refer indeed to handwriting_ocr.ipynb