Dimension of latent space
ask5898 opened this issue · 2 comments
This is more of a conceptual question. Generally VAEs are used to compress data into a lower dimensional latent space. However, in VAME, the input is of size Num_framesx12 and the latent dimension is (Num_frames-time_window)x30, which is a higher dimension. This is a bit counter-intuitive to the use of a VAE. Could you please clarify this?
thanks
Hi,
The input size for one time step is time_window x number_of_dlc_markers, given a window of 30 and 12 markers that would be 360, for example.
On the contrary, the latent space for one time step is as defined - default is 30 but can be as little as 10 in some cases (for simple patterns).
So the compression rate would be 360/30 = 12 for the default case ( and could he much higher).
Hope this helps
Best
Pavol
Oh yes that helps. Thanks for the clarification!!!