ValueError: You are passing a target array of shape (32, 1) while using as loss `categorical_crossentropy`. `categorical_crossentropy` expects targets to be binary matrices (1s and 0s) of shape (samples, classes). If your targets are integer classes, you can convert them to the expected format via: ``` from keras.utils import to_categorical y_binary = to_categorical(y_int) ``` about imageai HOT 11 CLOSED

Hi,

@allanis79. I'd suggest you to add a second category e.g. no-phone. (you'll have two categories: phone and no-phone)
You need to show to classifier how photo with phone looks like and how photo without phone looks like.
If you'll have just one label then classifier will assign all photos to this label.
EDIT: Probably that's why having one label is not allowed

Hopefully it helps you.

Best regards,
Tomek

from imageai.

thanks, I solved it. :)

from imageai.

So, I was getting the same issue here and I fixed changing the expected output type.

I was expecting an integer, after parse the integer to a plain-string, the error was gone.

from imageai.

can not train one object.

from imageai.

I am getting the same error, can I know how you solved it?

from imageai.

I am getting the same error, can I know how you solved it?

You must have at least 2 different types of object you need the model to recognize.

from imageai.

thanks for your quick reply.

see the above image? I want to detect the phone in that image. So I have only one label. Any ideas on how to approach the above problem?

from imageai.

@allanis79 , I guess you created the no-phone class, right? If that is the case, how many images did you need for this class - and, if applicable - did you do any comparison on the effect of the number of images for this class?

from imageai.

yes I did create a no phone class. I made around 700 images. I didn't compare it with anything else.

from imageai.

I know that this is old and may not be watched, but quick question. If I am doing something similar, basically it is or isn't there, do I need provide consistent images for the 'not there' option or can I provide random images? For the testing does it need to be anything consistent for the 'not there' option?

from imageai.

I think the more images you provide for the not-there class, the better it is. In your case I think it totally makes sense for them to be random and include as much different domains as possible

from imageai.