How to train the MeetRoom and VRU datasets? More detailed implementation should be provided.

[QZ1-boy]

The training and testing implemenation of MeetRoom and VRU datasets are missing in Readme file. They are should be provided for experiment replicability.

Besides, which model is the Swift4DLite and which model is Swift4D, sholud be clarified in this project.

Thank you very much!

[WuJH2001]

Hello, @QZ1-boy , the configuration of the MeetRoom dataset and the VRU dataset is the same as N3DV, with only the following differences:
For the MeetRoom dataset, the hash table size only needs to be $2^{15}$. For the larger VRU dataset, it is better to set it to 20. In addition, since the VRU dataset contains a basketball court scene that is relatively large and involves greater motion, you need to set the number of frames per training to 20 (N3DV and MeetRoom are 300 frames), while keeping other settings as default.

Regarding Swift4DLite and Swift, the only difference lies in the hash table size. For the N3DV dataset, the Lite version only requires setting the hash table size to $2^{16}$, with other settings unchanged. This makes training faster, but the quality will inevitably decrease to some extent.