How do I test the picture about keras-fcn

[JihongJu]

@pyziq Could you please provide your code, and training log?

@JihongJu
Thank you for reply

The code is:

`from keras_fcn import FCN
from keras.preprocessing import image
from keras.models import load_model
import numpy as np

model = load_model('fcn_vgg16.h5')
img = image.load_img('cat.jpg', target_size=(500, 500))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
preds =model.predict(x)
print preds`

and get the answer like this:

the traning logs:
I get the loss：1.92e-07 acc:0.7435 val_acc:0.7432
Is the loss too small?

[JihongJu]

@pyziq Hmm, that's interesting. Would you mind to share with me a copy of your trained model? Otherwise, it may take a few days for me to reproduce this bug. My email address is [email protected].

@JihongJu

I have sent the trained model.
Thank you very much!

[JihongJu]

@pyziq Sorry for the late reply. I just added a callback function to determine which layer the nans arise. It seems score_feat2 was the problem in my case. I am still debugging what was happening. Maybe you can also debug for what's happening in your case, with the new added keras_fcn.callbacks.CheckNumericsOps

Thank you for reply， I will try it

[JihongJu]

Hi @pyziq, it turns out that the problem was caused by the lack of an activation layer on top of the last output layer. Now it is fixed.