Dataset preprocessing

[LogicStuff]

Hello, I would like to know, why you have chosen to select only a subset of pedestrians in a scene, for which there are all self.seq_len points:

sgan/sgan/data/trajectories.py

Lines 131 to 132 in 691231e

	if pad_end - pad_front != self.seq_len:
	continue

This effectively disables padding and loss masking - pad_front is always 0 and pad_end is self.seq_len. Therefore, subsequences created in this manner may not (usually will not) contain all pedestrians. Is that desirable and superior to masking?

[WangHonghui123]

@LogicStuff I would like to know if you understand why to do so? If you know, could you explain it? Thank you.