Target $I$ for the Gated-LSTM baseline #63
Description
Hello, I have been trying to implement the Gated-LSTM baseline according to the descriptions in the RoomNav paper (link), but I am confused with the some of the notations.
From Section 5.2.2 (for Gated-LSTM):
The symbol $I$ is used to denote the target instruction. This same symbol was used in Section 5.1.2 (for Gated-CNN) to denote the target instruction prior to conversion to an embedding vector $y$:
So if we were to follow the same notation as in Section 5.1.2, we would have to feed the one-hot vector $I$ (and not the embedding vector) as input to the gated fusion module and the LSTM for the Gated-LSTM.
However, the Figure 2 seems to imply that it is in fact the embedding vector that is fed into the gated fusion and the LSTM:
Could you clarify whether the target $I$ for Gated-LSTM baseline is meant to be the one-hot vector or the embedding vector?
Thank you in advance.