search

chaneyddtt / Convolutional-Sequence-to-Sequence-Model-for-Human-Dynamics

Source code of paper "Convolutional Sequence to Sequence Model for Human Dynamics"
MIT License
86 stars 18 forks source link

Framerate for CMU data #4

Open jutanke opened 4 years ago

jutanke commented 4 years ago

I was wondering at what framerate your CMU data is sampled. Human3.6M is sampled at 50Hz and downsampled to 25Hz for training/evaluation but the original CMU dataset is sampled at 120Hz (and some sequences at 60Hz) which cannot easily be downsampled to 25Hz. Did you interpolate the data to acquire 25Hz for CMU? In the paper you state that

... We pre-process the (CMU-)data and evaluate the results in the same way as we did on the Human 3.6M dataset.

Could you elaborate if that means that the data is subsampled to 50Hz?

jutanke commented 4 years ago

I generated two sequences with 25Hz for each dataset and it seems to me that the CMU dataset is sampled at a slightly higher framerate: CMU (from this repo) cmueval Human3.6M h36m

chaneyddtt commented 4 years ago

Hi @jutanke, Sorry for the confusion. I used the script provided by @uoguelph-mlrg to convert the original data. The data has been downsampled to 30Hz as they described in their paper. I made a mistake in the table3 and table4 where the temporal horizon should be slight different from human3.6 dataset.

jutanke commented 4 years ago

Great, thanks for clearing this up!