search

Belval / CRNN

A TensorFlow implementation of https://github.com/bgshih/crnn
MIT License
299 stars 101 forks source link

What does seq_len means? #8

Closed PsyDog5hao closed 6 years ago

PsyDog5hao commented 6 years ago

Thank you for sharing this wonderful project. It works quite well for my number recg problem. But I have some confusion.

In feed_dict, self.__seq_len: [self.max_char_count] * self.data_manager.batch_size, and max_char_count = reshaped_cnn_output.get_shape().as_list()[1] . I'm not quite understand of that. Should seq_len be the width of img after cnn part of shape(batchsize, ?)

Sorry for my poor English.

Belval commented 6 years ago

After the CNN part of the model, the image will become a tensor of size 24x512 (if you use a width of 100px). The 24 represents the maximum number of characters that we could "read" and therefore the number of timesteps that the RNN should process.

RNNs are Recurrent Neural Networks that are fed an input for every step + the feed back from the previous step. seq_len (sequence_length) represents how many step we can provide.

I would recommend reading the paper as the author explains it quite well.

PsyDog5hao commented 6 years ago

3q