I was following this tutorial at https://www.tensorflow.org/versions/master/tutorials/image_retraining
After I went through a few hurdles, I got retrain.py running. Then I realized the new model was only able to classify images for the new categories I added. I am wondering if there is anything I did wrong here. I thought the new model should handle newly added categories in addition to the 1000 classes pre-trained for. Any input/comment is appreciated. Thanks.
The naive answer is: No you should'nt. Because the idea behind retrain is Transfer Learning.
According to Wikipedia
Transfer Learning focuses on storing knowledge gained while solving one problem and applying it to a different but related problem.
So source and target labels should be different.
But of course you can do what you want by combining source and target datasets and run the model by feeding the combined dataset. In your case training would take too much time, because inception dataset is not really small.
