Add classes to existing Inceptionv3 classes about deeplearningsandbox HOT 3 OPEN

Hi,

You can train Inception v3 to recognise classes not in ImageNet. You can train any pre trained network to recognise other classes.

1.Bottom conv blocks will act as general feature extractors. ( Frozen )
2.Upper Conv Blocks -- > Retrained with your dataset ( Unfrozen : Fine tuned to your dataset )
3.Fully connected layer --> Trained with your dataset.

You can just plug the same code as it is.

from deeplearningsandbox.

Thanks, I know I can train Inception v3 to recognise classes not in ImageNet.
My question was to make a model that recognizes classes in ImageNet and EVEN my own classes.
Basically I want to extend Inception v3 to recognize not only 1000 classes from ImageNet, even my 2 other classes. So in total it should recognizes 1002 classes.

from deeplearningsandbox.

from deeplearningsandbox.