Message Boards Message Boards

What Game of Thrones character do you resemble?

Posted 6 years ago
POSTED BY: Sam Tone
9 Replies

Sam, Thank you for an interesting post. If you don't have enough images, it might be a good idea to use the pre-trained model to build a classifier(Transfer Learning). There is an example in Wolfram Neural Net Repository. There are many examples besides ResNet-50 in Wolfram Neural Net Repository. I'm sorry I do not know which is better.

POSTED BY: Kotaro Okazaki

What are the steps for Transfer Learning in this case? Sam's net was trained just on a few classes of images which are exactly the GoT characters. ResNet-50 was trained on dataset "consisting of 1.2 million training images, with 1000 classes of objects." How do we squeeze 1000 classes of objects into just a few of GoT?

POSTED BY: Sam Carrettie

I understand Transfer Learning as follows.

? Adapt a model trained in an area where data can be widely applied to another area where data collection is difficult

? Lower levels of CNN seem to learn general-purpose features

? Apply to the new domain by learning only the part close to the output

So, the steps in this case are written in the section Transfer Learning of the example as follows.

? Remove the linear layer from the pre-trained net

? Create a new net composed of the pretrained net followed by a linear and softmax layer(LinearLayer[6] in this case)

? Train on the dataset, freezing all the weights except for new layer

POSTED BY: Kotaro Okazaki

Yes, Kotaro, thank you. I have also read about this recently. I also think ResNet-101 will be more useful as it is trained on faces.

POSTED BY: Sam Carrettie

This has been a super helpful conversation. Thanks for the input, Sam and Kotaro!

POSTED BY: Sam Tone

Hi Sander -- Thanks for the positive feedback! The list had 1429 images total. I think the breakdown was about 250 for Dany, 250 for Jon, 200 for Tyrion, 175 for Cersei, 150 for Jaime, 200 for Arya, and 225 for Sansa (obviously including in/out of character photos).

POSTED BY: Sam Tone

Thanks for sharing!How did you get the images? WebImageSearch? or manually? orÂ… ?

POSTED BY: Sander Huisman

I did them manually through drag and drop as to not use the WebImageSearch credits, which can be limited. It also made it a bit easier to pull images that I knew would probably pass my FindFaces iterator.

POSTED BY: Sam Tone

That's really neat! Thanks for sharing. How many faces did you use for learning?

POSTED BY: Sander Huisman
Reply to this discussion
Community posts can be styled and formatted using the Markdown syntax.
Reply Preview
Attachments
Remove
or Discard

Group Abstract Group Abstract