【OCR】ocr改shape

最新推荐文章于 2022-10-24 21:20:22 发布

Mein_Augenstern

最新推荐文章于 2022-10-24 21:20:22 发布

阅读量198

点赞数

分类专栏： OCR

本文链接：https://blog.csdn.net/Maisie_Nan/article/details/86715573

版权

OCR 专栏收录该内容

8 篇文章 0 订阅

订阅专栏

reader里datashape从[160,32]改为[640,32]，则

class CRNN:
    def __init__(self, data, label, rnnSeqLengths, rnnBatch, conf, isTraining, keepProb, reuse):
        self.data = data
        self.label = label
        self.rnnSeqLengths = rnnSeqLengths
        self.rnnBatch = rnnBatch
        self.conf = conf
        self.isTraining = isTraining
        self.keepProb = keepProb
        net = DenseNet(self.data, 2, 12, self.isTraining, dropout=1-self.keepProb, reuse=reuse).model
        net = tf.transpose(net, perm=[0, 2, 1, 3])
        # self.word_vec = tf.reshape(net, [1, -1, 1584])
        self.word_vec = tf.reshape(net, [-1, 20, 816])
        #print 'self.word_vec', self.word_vec.shape
        self.lstmLayers(reuse=reuse)

改为：

class CRNN:
    def __init__(self, data, label, rnnSeqLengths, rnnBatch, conf, isTraining, keepProb, reuse):
        self.data = data
        self.label = label
        self.rnnSeqLengths = rnnSeqLengths
        self.rnnBatch = rnnBatch
        self.conf = conf
        self.isTraining = isTraining
        self.keepProb = keepProb
        net = DenseNet(self.data, 2, 12, self.isTraining, dropout=1-self.keepProb, reuse=reuse).model
        net = tf.transpose(net, perm=[0, 2, 1, 3])
        # self.word_vec = tf.reshape(net, [1, -1, 1584])
        self.word_vec = tf.reshape(net, [-1, 80, 816])
        #print 'self.word_vec', self.word_vec.shape
        self.lstmLayers(reuse=reuse)