why the output probability of caffe's softmax layer is [1, 0, 0....], only one element is 1

[Qingze Wang]

I train my own model to classify characters, accuracy 98%, but when I use the example in examples/cpp_classification/, output probability of caffe's softmax layer is [1, 0, 0....], only one element is 1,  others are 0, why? 

[Qingze Wang]

I did not subtract the mean, it is the problem?

在 2015年12月7日星期一 UTC+8下午1:37:38，Qingze Wang写道：

[Felix Abecassis]

It means the confidence of the first class is 100%. What's the problem?

[Qingze Wang]

but it's not the right label.

I am using caffe to classify characters, one stage of ocr, the problem is that on the stage of segmentation I am not sure all croped images is right, i.e. all croped images has a character in it, maybe it's a noise images, but when  I give the noise image to caffe classifier it gives me a characeter label, so this is wrong.

my idea is that if the softmax of caffe has a good probability distribution, I can use this to help segmentation.

在 2015年12月8日星期二 UTC+8上午2:26:19，Felix Abecassis写道：