convert to model to tensorRT

[YaelGrinstein]

Hi, Thanks for sharing the code with your fundamental work. I would like to convert the Prototype Net to tensorRT or ONNX, do you have any experience with that? I will appreciate any clues. best regards, Yael

[hushell]

Hi Yael, for on-device deployment, you'll need a static graph, which means you'll have more inputs than a normal image classification network. The forward pass will be something like this:

output = model(supp_x, supp_y, query)

I didn't try myself but Samsung engineers did manage to run a demo on Android.