Difference between the `plot` and `plot_rec` functions

[lpcinelli]

Hi, thank you for making your work available. I have a few questions regarding the plot functions.

• What is the main difference between the plot and plot_rec functions?

  • plot_rec conditions on groundtruth frames throughout the whole sequence, while plot only during the initial context phase (prior to opt.n_past) after which conditions on its on predictions. Is it correct?
  • Is there any special reason for plot_rec generating frames in range(1, opt.n_past+opt.n_future) a while plot in range(1, opt.n_eval)?

• The plot function in file train_svg_fp.py does not have the following line

  posterior.hidden = posterior.init_hidden() 

  while all other plot functions (both svg_fb and svg_lp files) have it. Is it on purpose or a mistake?

Thanks again :)

[gitzyong812]

In real generation situation, we only have some initial frames to generate the next. And we don't have ground truth frame to infer the posterior. This corresponds to the 'plot' function.
But the 'plot_rec' function uses ground truth frame to get posterior.