Torch and list data as learning features

1

Hi,
i need some explanation about list data ,
here I am using transfer learning features(fc) suppose 1024 features(saved into pickle file). i have saved all features and later imported them into fearure=[] (as list data)

for img in features:
    img=torch.FloatTensor(img)
    output=model(img)

do I need set img.requires_grad=True, before feed to the model?

2

Usually you wouldn’t need to set requires_grad=True on the inputs unless you explicitly want to update the inputs for e.g. an adversarial use case.

3

Ok. Thanks. I just need to put them in as “the thought vector” in LSTM.