Possible discrepancy between the reported result and the F2GAN paper

[IssamLaradji]

Thanks for this great code, it's a relief to find a pytorch version of DAGAN.

The README reported accuracy on 5-shot omniglot is 69% accuracy using DAGAN which is widely different from the 88.81% accuracy reported in Table 2 in the F2GAN paper (https://arxiv.org/pdf/2008.01999.pdf). Could you help me find out why there is a huge difference in the results? It seemed like your problem setup and the one in F2GAN are the same.