yuqicui / mmd_aae

Code refactoring for paper "Domain Generalization with Adversarial Feature Learning" in VLCS datasets.

MMD_AAE

Implementation of Domain Generalization with Adversarial Feature Learning

How to use?

Download Decaf features from VLCS datasets and put the mat file in data/VLCS. If you have different file format, you can modify the load_data function in utils.datasets. Then use the following command:

python3 train.py

mmd_aae's Issues

Question about the loss

It seems the order of losses are wrong.

Where to get the data?

Hi,

Thank you for sharing the implementation. Could you also provide guidance of how to obtain the data? Or simply upload the data? Thanks!

Questions with respect to the code

Hi, @YuqiCui , Thanks for your implementation. I have a question that if I change the name of encoder and add it like "model = Sequential(name='encoder')", error will occur like:

ValueError: Unknown entry in loss dictionary: "encoder". Only expected the following keys: ['decoder', 'dense_1', 'adv', 'task']

However, it is a serious problem since if I wanna multiple conv layers in the encoder, the loss as a whole cannot be computed. Hopfully you can help me. :)

Thanks,