First of all, thank you very much for a large amount of processed data. There are no problems for research purposes, but there are some questions about the dataset in competition.
In test dataset 2, 13, 75, etc., we can see that the time difference between frames is not constant. In the paper, the time difference between frames of the dataset is 10hz, but the difference between the actual dataset is 6hz or less.
I think the noise from the observation dataset is helpful for robust learning, but I think the exact time interval(10Hz) is important to compare the predicted trajectory and ground truths. (The error between generating 30 frames of trajectory at 10 Hz and generating 6 Hz is quite large)
I think it is impossible to predict the frame’s interval of data when t> 2s of the test dataset. To solve this, what is the time difference between frames in the test dataset when t> 2s? Is it fixed or can you provide it?