shawnh2 / bankcard-recognizer Goto Github PK

在模型里面的的参数 input(None,32,256,1) 32和256是输入图像的像素值么。32*256

你好！ 我想请问下为什么修改batch size之后报tensorflow.python.framework.errors_impl.InvalidArgumentError: Not enough time for target transition sequence (required: 4, available: 3)1You can turn this error into a warning by using the flag ignore_longer_outputs_than_inputs
这个错误？
谢谢

卡号定位是不是非得用完整的银行卡图片才能做成？无论是完整大小的银行卡图片和裁剪过的我都找不到。

参数：
AUG_NBR = 100

11262/11262 [==============================] - 4133s 367ms/step - loss: 3.9718 - acc: 1.1099e-05 - val_loss: 2.8851 - val_acc: 0.0000e+00
11262/11262 [==============================] - 4147s 368ms/step - loss: 2.8700 - acc: 1.1099e-05 - val_loss: 2.8474 - val_acc: 0.0000e+00
11262/11262 [==============================] - 4099s 364ms/step - loss: 2.8316 - acc: 0.0000e+00 - val_loss: 2.8033 - val_acc: 0.0000e+00
11262/11262 [==============================] - 4043s 359ms/step - loss: 2.8115 - acc: 2.2199e-05 - val_loss: 2.7982 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3998s 355ms/step - loss: 2.8044 - acc: 0.0000e+00 - val_loss: 2.7776 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3987s 354ms/step - loss: 2.7995 - acc: 0.0000e+00 - val_loss: 2.7764 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3989s 354ms/step - loss: 2.7964 - acc: 0.0000e+00 - val_loss: 2.7756 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3990s 354ms/step - loss: 2.7916 - acc: 3.3298e-05 - val_loss: 2.8062 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3991s 354ms/step - loss: 2.7950 - acc: 0.0000e+00 - val_loss: 2.7778 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3989s 354ms/step - loss: 2.7866 - acc: 1.1099e-05 - val_loss: 2.8157 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3990s 354ms/step - loss: 2.7643 - acc: 0.0000e+00 - val_loss: 2.7624 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3987s 354ms/step - loss: 2.7613 - acc: 0.0000e+00 - val_loss: 2.7621 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3986s 354ms/step - loss: 2.7620 - acc: 0.0000e+00 - val_loss: 2.7604 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3988s 354ms/step - loss: 2.7616 - acc: 0.0000e+00 - val_loss: 2.7653 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3988s 354ms/step - loss: 2.7613 - acc: 0.0000e+00 - val_loss: 2.7633 - val_acc: 0.0000e+00
11262/11262 [==============================] - 3991s 354ms/step - loss: 2.7617 - acc: 0.0000e+00 - val_loss: 2.7618 - val_acc: 0.0000e+00
11262/11262 [==============================] - 4154s 369ms/step - loss: 2.7606 - acc: 0.0000e+00 - val_loss: 2.7605 - val_acc: 0.0000e+00
11262/11262 [==============================] - 4092s 363ms/step - loss: 2.7599 - acc: 0.0000e+00 - val_loss: 2.7601 - val_acc: 0.0000e+00
11262/11262 [==============================] - 4089s 363ms/step - loss: 2.7602 - acc: 0.0000e+00 - val_loss: 2.7608 - val_acc: 0.0000e+00
11262/11262 [==============================] - 4079s 362ms/step - loss: 2.7597 - acc: 0.0000e+00 - val_loss: 2.7604 - val_acc: 0.0000e+00

感谢楼主的分享，看到您当前提交的程序中已经对之前其他同学提出的类似问题做出了修改，但是我这边跑训练时候，CRNN模型仍然在降低到2.7-2.8附近后，不再下降，得到的模型与楼主分享的相差较大，希望您看到后能给予指导，谢谢。

res.append(int(ch))

ValueError: invalid literal for int() with base 10: 'i'
0%| | 0/548 [00:00<?, ?it/s]
这里会报错怎么不办

请问您的识别模型当初训练了多久呀，我基于预训练的参数，loss最低降到2.6左右，之后上升。最终early stop。
训练有什么技巧吗。
谢谢。

您好，我想请问一下，您项目里面的east模型训练到最后平均loss是多少