请问作者训练了多少个epoch之后，sequence_accuracy开始有值？ about crnn.tf2 HOT 4 CLOSED

       把本batch的图补齐等长
        for i in range(len(train_image_data)):
            template = np.zeros((32, batch_max_width, 3))
            template[:32, :train_image_data[i].shape[1], :] = train_image_data[i]
            train_image_data[i] = template
        train_image_data = np.array(train_image_data)

        把本batch的label补齐等长
        for i in range(len(labels)):
            template = np.zeros(batch_max_label_length)  
            template[: len(labels[i]) ] = labels[i]
            labels[i] = template
        labels = np.array(labels) 

 def call(self, y_true, y_pred):
        y_true = tf.cast(y_true, tf.int32)
        logit_length = tf.fill([tf.shape(y_pred)[0]], tf.shape(y_pred)[1])
        label_length = tf.fill([tf.shape(y_true)[0]], tf.shape(y_true)[1])
        loss = tf.nn.ctc_loss(
            labels=y_true,
            logits=y_pred,
            label_length=label_length,
            logit_length=logit_length,
            logits_time_major=False,
            blank_index=0)
        return tf.reduce_mean(loss) 

def update_state(self, y_true, y_pred, sample_weight=None):
        y_true_shape = tf.shape(y_true)
        batch_size = y_true_shape[0]
        y_pred_shape = tf.shape(y_pred)
        max_width = tf.maximum(y_true_shape[1], y_pred_shape[1])
        logit_length = tf.fill([batch_size], y_pred_shape[1])      
        decoded, _ = tf.nn.ctc_greedy_decoder(
            inputs=tf.transpose(y_pred, perm=[1, 0, 2]),
            sequence_length=logit_length)
        y_pred = self.to_dense(decoded[0], [batch_size, max_width])
         仅这里有变化，我觉得我对您的-1理解有欠缺，所以在这里用from_dense把补的0抹掉了，变稀疏矩阵，再用您的to_dense，也就是用-1补齐了，恢复成您源码中输入给tf.math.reduce_any的数据原貌。
        y_true = tf.sparse.from_dense(y_true)
        y_true = self.to_dense(y_true, [batch_size, max_width])
        num_errors = tf.math.reduce_any(
            tf.math.not_equal(y_true, y_pred), axis=1)
        num_errors = tf.cast(num_errors, tf.float32)
        num_errors = tf.reduce_sum(num_errors)
        batch_size = tf.cast(batch_size, tf.float32)
        self.total.assign_add(batch_size)
        self.count.assign_add(batch_size - num_errors)