Possible bug in downsampling

[abhinanda-sony]

Hi!

Thank you for this awesome work! After looking at the data processing steps, I think there is a bug in the way motions are downsampled.

In line 183 of the file raw_pose_processing.ipynb, the downsampling factor is calculated as follows:

HumanML3D/raw_pose_processing.ipynb

Line 183 in 9176e8f

" down_sample = int(fps / ex_fps)\n",

This works well for most of AMASS, but for sequences from SSM_synced, the fps of the sequences is 59.9999885559082.
The downsample factor then becomes 2, but it should be 3.
Thirteen sequences (not including mirrored sequences) from SSM_synced have this incorrect downsampling.

This also affects sequences from the Eyes_Japan_Dataset, which have an original FPS of 250 in AMASS because the downsampling is not done evenly.
There are 129 sequences with this problem from SSM_synced and Eyes_Japan_Dataset.