Toqi Tahamid

You need to exclude the logistic layer of inception as ImageNet has 1000 class and add your own logistic layer. Here is a example - https://github.com/fchollet/keras/blob/master/docs/templates/applications.md#fine-tune-inceptionv3-on-a-new-set-of-classes

What are you doing in this method decode_predictions(preds)? Can you share the code?