zhengwh/captcha-tensorflow

hack captcha by cnn with tensorflow

captcha-tensorflow's Issues

数据集大小

请问一下，你的数据集具体给了多大？谢谢

如果识别jpg格式的识别码，出现错误

错误提示信息：
'JpegImageFile' object has no attribute 'shape'

loss function的优化

如果最终要预测的是四个字符，把loss定义为四部分交叉熵loss的平方和（每部分label是一个one-hot向量），效果会显著提升，现在这样直接把输出结果和一个four-hot向量做交叉熵，意义不大

ValueError: could not broadcast input array from shape (9600) into shape (4000)

您好，我的数据集大小为 40*100，然后使用本项目进行训练的时候，在

batch_x[i, :] = image.flatten() / 255 # (image.flatten()-128)/128 mean为0

这一行抛错了，错误为

ValueError: could not broadcast input array from shape (9600) into shape (4000)

请问这个应该如何解决呢？或者能提提思路吗？

感谢

训练时修改图片大小报错的问题

训练时把生成的图片高改成了70就报错了，错误如下:
Caused by op 'my_monitor/softmax_cross_entropy_with_logits_sg', defined at:
File "D:/workspace-python/captcha-tensorflow-master/capt/train.py", line 104, in
train_crack_captcha_cnn()
File "D:/workspace-python/captcha-tensorflow-master/capt/train.py", line 26, in train_crack_captcha_cnn
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=predict, labels=label))
File "D:\softs\Anaconda3\envs\app_002\lib\site-packages\tensorflow\python\util\deprecation.py", line 250, in new_func
return func(*args, **kwargs)
File "D:\softs\Anaconda3\envs\app_002\lib\site-packages\tensorflow\python\ops\nn_ops.py", line 1959, in softmax_cross_entropy_with_logits
labels=labels, logits=logits, dim=dim, name=name)
File "D:\softs\Anaconda3\envs\app_002\lib\site-packages\tensorflow\python\ops\nn_ops.py", line 1873, in softmax_cross_entropy_with_logits_v2
precise_logits, labels, name=name)
File "D:\softs\Anaconda3\envs\app_002\lib\site-packages\tensorflow\python\ops\gen_nn_ops.py", line 7702, in softmax_cross_entropy_with_logits
name=name)
File "D:\softs\Anaconda3\envs\app_002\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 787, in _apply_op_helper
op_def=op_def)
File "D:\softs\Anaconda3\envs\app_002\lib\site-packages\tensorflow\python\framework\ops.py", line 3392, in create_op
op_def=op_def)
File "D:\softs\Anaconda3\envs\app_002\lib\site-packages\tensorflow\python\framework\ops.py", line 1718, in init
self._traceback = self._graph._extract_stack() # pylint: disable=protected-access

InvalidArgumentError (see above for traceback): logits and labels must be broadcastable: logits_size=[432,36] labels_size=[384,36]
[[Node: my_monitor/softmax_cross_entropy_with_logits_sg = SoftmaxCrossEntropyWithLogits[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"](my_monitor/softmax_cross_entropy_with_logits_sg/Reshape, my_monitor/softmax_cross_entropy_with_logits_sg/Reshape_1)]]

softmax_cross_entropy_with_logits和sigmoid_cross_entropy_with_logits的区别

我做的是单个字符的验证码识别，用softmax_cross_entropy_with_logits计算loss时，loss降到3.5左右就不再变化，改用sigmoid_cross_entropy_with_logits后loss会下降到0.15左右就不再下降，这是什么原因呢？