我训练了20轮
3.基于LSTM+CTC实现不定长文本图片OCR上一篇实现了图片CNN多标签分类(4位定长验证码识别任务)(地址:https://www.jianshu.com/p/596db72a7e00) 本文继续优化,实现不定长文本...
self.words = '0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTVUWXYZ'
for idx, val_seq in enumerate(val_sequence):
val_pred_tmp = [self.words.find(x) if self.words.find(x) > -1 else 62 for x in val_seq]
val_y_true_tmp = [x for x in y_val_tmp_true[idx] if x != 62]
我把26修改成62了,然后数字字母加在一起就这样了,还是训练次数少的原因吗
3.基于LSTM+CTC实现不定长文本图片OCR上一篇实现了图片CNN多标签分类(4位定长验证码识别任务)(地址:https://www.jianshu.com/p/596db72a7e00) 本文继续优化,实现不定长文本...
@周小叨 没有报错识别效果不好,false[]里面现在都是空,
False [] <<==>> ['4', 'l', 'D']
False [] <<==>> ['7', 'h']
False [] <<==>> ['y', 'j', 'g']
False [] <<==>> ['b']
False [] <<==>> ['f']
False [] <<==>> ['P', 'C', '8']
False [] <<==>> ['D', 'X']
False [] <<==>> ['q', 'x']
False [] <<==>> ['Q']
False [] <<==>> ['d', 'j', 'P', 'W']
False [] <<==>> ['J', 'E', 'w']
False [] <<==>> ['J', 'a', 'a']
False [] <<==>> ['Q', 'I', 'G', 'Q']
False [] <<==>> ['r', 'v', 'A', 'B']
False [] <<==>> ['e', 'h']
False [] <<==>> ['P', '8', '6', 'j']
False [] <<==>> ['h']
3.基于LSTM+CTC实现不定长文本图片OCR上一篇实现了图片CNN多标签分类(4位定长验证码识别任务)(地址:https://www.jianshu.com/p/596db72a7e00) 本文继续优化,实现不定长文本...
我单独训练数字,字母都没有 问题,为啥加一起就不可以了
3.基于LSTM+CTC实现不定长文本图片OCR上一篇实现了图片CNN多标签分类(4位定长验证码识别任务)(地址:https://www.jianshu.com/p/596db72a7e00) 本文继续优化,实现不定长文本...
@周小叨 谢谢,我多跑一会看下
3.基于LSTM+CTC实现不定长文本图片OCR上一篇实现了图片CNN多标签分类(4位定长验证码识别任务)(地址:https://www.jianshu.com/p/596db72a7e00) 本文继续优化,实现不定长文本...
'E', 'C', 'E', 'C', 'E', 'B', 'E', 'G', 'F', 'S', 'P', 'I', 'P', 'I', 'E', 'I', 'F', 'C', 'X', 'G', 'A', 'O', 'C', 'A', 'V', 'O', 'B', 'E', 'I', 'E', 'I', 'E', 'C', 'O', 'S', 'C', 'O', 'I', 'F', 'I', 'F', 'X', 'E', 'C', 'E', 'V', 'X', 'V', 'E', 'X', 'E', 'X', 'E', 'X', 'C', 'A', 'C', 'A', 'V', 'J', 'Y', 'M', 'F', 'E', 'V', 'E', 'J', 'E', 'Q', 'E', 'T', 'E', 'V', 'E', 'V', 'E', 'T', 'E', 'Q', 'E', 'A', 'E', 'Q', 'E'] <<==>> ['M', 'L', 'D', 'J', 'S']
train_loss:687.7257690430 train_distance:11.3125000000
val_loss:685.5680541992 val_distance:11.2533330917 val_acc:0.0000000000
epoch:0/200 batch:100/595 total_step:100 lr:0.0000500000
False ['G', 'F', 'S', 'V', 'R', 'F', 'G'] <<==>> ['L', 'Y', 'P', 'Q']
False [] <<==>> ['P', 'B']
False [] <<==>> ['V', 'T', 'W', 'H', 'F', 'Z', 'E', 'I', 'A']
False [] <<==>> ['V', 'K', 'H', 'G', 'J', 'I', 'J', 'B', 'E', 'I']
False ['F'] <<==>> ['T', 'S', 'Y', 'T', 'L']
train_loss:30.3770618439 train_distance:0.9937499762
val_loss:31.3891277313 val_distance:1.0041667223 val_acc:0.0000000000
epoch:0/200 batch:200/595 total_step:200 lr:0.0000500000
第一批次和第二批次打印出来不一样, 训练时出现的情况, 无法准确验证。
3.基于LSTM+CTC实现不定长文本图片OCR上一篇实现了图片CNN多标签分类(4位定长验证码识别任务)(地址:https://www.jianshu.com/p/596db72a7e00) 本文继续优化,实现不定长文本...