When Text2MotionDataset loads cached files (in tmp folder), length_list (self.length_arr) is not properly recreated.

[xaiseung]

In mGPT/data/humanml/dataset_t2m.py, line 65~85:

new_name_list = []
length_list = []
data_dict = {}

# Fast loading
...
    with open(...) as file:
        name_list = pickle.load(file)
    for name in new_name_list:
        length_list.append(data_dict[name]['length'])

The cached list of names is stored as name_list, but in line 84, new_name_list (an empty list) is used.
This causes a problem where selt.reset_max_len() does not work as intended.

proposed fix

This can be resolve by:

    for name in name_list:
        length_list.append(data_dict[name]['length'])

I know the original code is from here, but the caching part seems to be added separetely, so I'm issuing this here.