Search code examples
python-3.xkerasvgg-netimagenet

How to find the parent class of VGG16 prediction?

I am using a keras based vgg model as an image classifier. From this webpage:

https://machinelearningmastery.com/use-pre-trained-vgg-model-classify-objects-photographs/

But I am interested to find out the parent category of a prediction. For example, if the model predicts a dog, I would like to know its parent category which is animal. I do not know if there is a way to use the Imagenet tree for this problem?

Solution

  • ImageNet uses WordNet to create the hierarchy of classes. You can access the object synset in the following way:

    import nltk
from nltk.corpus import wordnet

nltk.download('wordnet')
dog = wordnet.synsets('dog')[0]

    To access the parents of the dog, call:

    dog.hypernyms()

    It might return more than 1 result which is problematic.

    What you can do now, is to check if any of this parents is an actual ImageNet class. To get the wnid of the symset, just call dog.offset().

    To reverse the process:

    from nltk.corpus import wordnet as wn
wn.synset_from_pos_and_offset('n',2084071)