Request training/evaluation dataset

[dlwjdgns3415]

Hi, Thanks for your open source and great works.
Can I get the University of Maryland (UMD) training/evaluation dataset instead of sample data to reproduce the same results as the experimental results in the paper?
I'm very confused by the different data structure of the sample data and MTG.

[Robot-SXY]

Hello, may I ask if you have solved this problem?

[dlwjdgns3415]

Unfortunately, I haven't been able to resolve it either.

[jingGM]

Sorry for the late response. The dataset for DTG is based on GND. For training purpose, we augmented the dataset by generating ground truth trajectories with A*, as mentioned in the paper. However, the process requires substantial manual effort to remove inconsistent observation–trajectory pairs. Because of this, we don’t have a fully automated pipeline for generating training and evaluation datasets from GND. Therefore, we decided not to release the entire benchmark of GND for now, but only release a sample dataset to help users understand the code and run the algorithm. To get started with the training and evaluation process with GND, I would suggest directly segmenting the rosbags to generate the observation–trajectory pairs.