Train/test split definition

[yizhou-wang]

Hi! Thanks a lot for your contribution to this valuable dataset! I have one question on the train/test split for this dataset. What should be a proper splitting method and make it comparable with the results in your paper (Table 4)?

[M-Eng]

Hi,

thank you for your interest. We are currently planning to update the paper with more recent numbers. The current results in table 4 were obtained with only partial data.

In the final version of the paper we use the second second half of the first sequence and only include the first 8 camera. For training we use the first half.

I have attached below an updated table with the most recent number for the multiview case:

We do so as this is the data with the cleanest annotation. The rest of the cameras/sequences can also potentially be used for training/eval but the annotations where automatically generated.