Can transfer learning be performed from the pre-trained image caption model

[kevjp]

From image captioning tutorial is there any way to continue training from the pre-trained model? If so how do we load the model's state_dict and the optimizer's state_dict? As far as i can make out I am unable to load these attributes from the pickle files decoder-5-3000.pkl and encoder-5-3000.pkl files.

So I can't perform the following:
checkpoint = torch.load('decoder-5-3000.pkl')
model.load_state_dict(checkpoint['model_state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])

Any help would be much appreciated.

[wys837]

good afternoon, can you share the pretrained.zip? Thanks a lot