MatchingNetworks

MatchingNetworks copied to clipboard

Cosine distance calculation problem

2

In the source code, the author calculates the cosine distance as follows. sum_support = torch.sum(torch.pow(support_image, 2), 1) support_manitude = sum_support.clamp(eps, float("inf")).rsqrt() dot_product = input_image.unsqueeze(1).bmm(support_image.unsqueeze(2)).squeeze() cosine_similarity = dot_product * support_manitude *...

ouc-lq

Thanks for your great work! I have a problem about how to run the code on the dataset miniImagenet. Does it mean that I should download full dataset "ILSVRC2012_img_train" and...

xinobu

Problem of the “full context embeddings” implement

3

First thank you for your code. But I am puzzled about your “full context embeddings” implementation. In the paper, the process of g' and f' are different, but you just...

cultivater

low accuracy at miniImagenet dataset?

7

First very thanks for your implement of Matching-Networks with pytorch. I have follow your setup to run the miniImagenet example,the training accuracy can achieve about 100%,but the val and test...

BoyuanJiang

The training objective in the original paper is theta. Should the loss value be theta? Why is loss in your program a cross entropy loss?

lwzhaojun

On [this line](https://github.com/gitabcworld/MatchingNetworks/blob/master/models/AttentionalClassify.py#L28) you're applying a softmax to the similarities. Then [later](https://github.com/gitabcworld/MatchingNetworks/blob/master/models/MatchingNetwork.py#L88) you apply `cross_entropy`, which is a log softmax + NLL loss. I think you probably want to remove...

bkj

lori0726

Thank you for sharing your code. I have a question here about the Omniglot dataset test, where the support set is coded with a convolutional neural network, followed by the...

Hjwinging

I have follow your setup to run the Omniglot example,test accuracy is about 99.6%.In origin paper it's about 98.9%.So I wonder if where I'm wrong to run your code or...

554374706