potential problem in code

[xweiyuan]

I think t_step = np.random.randint(max_t_step) + 1 in ./data/base_dataset.py would throw an error when cur_seq_len equals to 1.

Maybe an if-clause like if max_t_step==0: t_step=1 should be added to avoid crash while training?

[tcwang0509]

cur_seq_len shouldn't really be 1, which means the video only consists of one frame. But I guess there's no hurt checking that too. Thanks for the info!

[xweiyuan]

Thanks for your reply and great work.
Besides, the code will throw out dimension mismatching error while fetching data as bellow:

'invalid argument 0: Sizes of tensors must match except in dimension 0. Got 4 and 3 in dimension 1 at /opt/conda/conda-bld/pytorch_1573049306803/work/aten/src/TH/generic/THTensor.cpp:689'

Maybe it is better to filter out the video sequences which is too short, like less than the seq_len_max ?