Seonghyeon Nam

@rookie-zzw I'm sorry but I can't reproduce your error. If you can use pdb or something for debugging, printing any suspicious variables might be helpful to find the culprit.

@MapleSpirit It is actually mentioned in the original paper. The authors used the term 'text embedding augmentation', which is same as the conditioning augmentation in StackGAN. Also, I personally got...

Hi! Wrapping sorted_indices with LongTensors might solve the problem as the error message suggested. Maybe you are using a different version of pytorch.

0.2 as described in README.

I'm sorry that I do not know the reason.. But, at least it worked at that time. Maybe you can try `torch.stack([vgg_normalize(i) for i in img.data])` instead `vgg_normalize(img.data)`.

Yes, I just used the annotated dataset and I didn't modify it except visual-semantic embedding...

@tangjialiang-jj Please visit [this website](https://fasttext.cc/docs/en/pretrained-vectors.html).

@Ereebay To me, they are convolutional layers. Can you double check your network? ``` >>> import torchvision.models as models >>> vgg = models.vgg16_bn() >>> vgg VGG( (features): Sequential( (0): Conv2d(3,...

@Ereebay I see. You are right. Can you submit a pull request for this? Thanks!

@Ereebay It may be natural due to the difference of gradient in early stage. But, the final result will be ok, I guess.