Asking why support set images are fed into convolutional neural networks and the role played by bi-LSTM？

[Hjwinging]

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 convolutional coding of the target dataset images. At this time I output the results and find that the support set images have specific values, while the target dataset images are 0, -1, 2, etc. with 0 after the decimal point. and the embedding vectors are fed into the bi-LSTM afterwards, I don't see what the bi-LSTM does.