INNER CODE UNIT · Python

encoder_out

graviraja/100-Days-of-NLP · applications/generation/image_captioning/latex_app/app.py:97

        encoder_out = encoded_image.view(batch_size, -1, encoder_dim)
        # encoder_out => [1, 16, 512]
        
        num_pixels = encoder_out.size(1)

        hidden, cell = model.init_hidden_state(encoder_out)
        # hidden, cell => [1, decoder_dim]

        hidden = hidden.unsqueeze(0)
        cell = cell.unsqueeze(0)
        # hidden, cell => [1, 1, decoder_dim]

        dec_inp = torch.LongTensor([vocab('<s>')]).to(device)
        sampled_ids = []

        for t in range(1, max_len):
            output, hidden, cell, _ = model.decoder(dec_inp.unsqueeze(1), hidden, cell, encoder_out)
            top1 = output.argmax(1)

View source record →