about concat

[chencodeX]

https://arxiv.org/pdf/1506.04214.pdf
in this parper

the LSTM compute i(t),f(t),o(t) is use x(t),h(t-1) and c(t-1). but in your code convolution_lstm.py line 24:
combined = torch.cat((input, h), dim=1)

you just use x and h, why?

[automan000]

Hi Chen,
I think you might be right. I forget why I did this when I wrote the code.
I have checked several implementations. Many of them make the same decision.
I will do some further investigations. Once finished, I will let you know.
Thanks anyway.

[chencodeX]

I have some of my own opinions, which may be to save code complexity and the number of parameters

[Jackie-Chou]

@automan000 Guess you have referenced some implementations of LSTM instead of convLSTM, note the difference of formulas of them .
LSTM

convLSTM

The raw LSTM really doesn't involve c for computing the gate while convLSTM does.

[automan000]

@Jackie-Chou Actually I am referring the implementations such as BasicConvLSTMCell
But your words are very convincing. I will update the code ASAP. Thanks a lot.

[automan000]

Bug fixed. Close this issue.