Hi there,
Is torch’s (not PyTorch) TemporalSubSampling, simply equivalent to PyTorch’s Conv1d?
Applies a 1D sub-sampling over an input sequence composed of
nInputFrame
frames. Theinput
tensor inforward(input)
is expected to be a 2D tensor (nInputFrame x inputFrameSize
). The output frame size will be the same as the input one (inputFrameSize
).
The output value of the layer can be precisely described as:
output[t][i] = bias[i] + weight[i] * sum_{k=1}^kW input[dW*(t-1)+k][i]
See TemporalSubSampling’s Lua code here.
But they have TemporalConvolution, from the description it looks more like PyTorch’s Conv1d, so I’m confuse what is TemporalSubSampling in PyTorch.