Missing indoor ground truth data?

[nstathou]

Hey, thanks for publishing this dataset.

I tried using the indoor data from the Laboratory sequences, however it seems that the files with the ground truth data only contain two instances, as seen in the attached image.

Is that supposed to be like that?

[Mechazo11]

Hi @nstathou I just looked and saw that all Laboratory scenarios have no ground-truth valid ground-truth data. @DapengFeng can you kindly uploaded the correct version?

[DapengFeng]

For the Laboratory sequences, we only record the start and end points, since the limited vicon devices.

[Mechazo11]

Hi @DapengFeng thank you for explaining the issue.

I just went over the S3E paper again and just saw the following

In indoor environments without GNSS signals, like laboratories, a motion capture system with 17 high-frequency cameras is used to record track start and endpoints due to the high costs of a large-scale system

May I suggest updating the README.md of the HuggingFace repo to mention this issue with the indoor datasets.

However, I am confused by something else. Given there was no expert system to give Absolute positions of the three robots, how were you able to evaluate the results below in the S3E paper

and in the CoLRIO paper (Table II)

[CYWCC]

Hi @DapengFeng thank you for explaining the issue.

I just went over the S3E paper again and just saw the following

In indoor environments without GNSS signals, like laboratories, a motion capture system with 17 high-frequency cameras is used to record track start and endpoints due to the high costs of a large-scale system

May I suggest updating the README.md of the HuggingFace repo to mention this issue with the indoor datasets.

However, I am confused by something else. Given there was no expert system to give Absolute positions of the three robots, how were you able to evaluate the results below in the S3E paper

and in the CoLRIO paper (Table II)

I have the same question. Were only the ​​first frame and last frame positions​​ used for evaluation? @DapengFeng

[DapengFeng]

@Mechazo11 @CYWCC Thank you for your suggestion. For experimental sequences in the laboratory, we capture only the start and end points due to constraints in Vicon system availability. Evaluation of these sequences is subsequently performed using only these two reference points. I've updated the HuggingFace repository's README.md documentation to reflect this methodological constraint and ensure transparency.