Skip to content

Latest commit

 

History

History
35 lines (18 loc) · 1.27 KB

dataset_preparation.md

File metadata and controls

35 lines (18 loc) · 1.27 KB

Dataset Preparation

For each dataset, you should first process the reconstructed dataset, which is the dataset without the -SV or -MV suffix. This dataset is used for model evaluation.

If you want to train the entire model, you need to further process the SV and MV data; if you only want to perform evaluation, you only need to process the MV data.

In addition to Tsinghua Cloud, we also upload the checkpoints and processed data to HuggingFace. Click here for more details.

ScanNet200:

For ScanNet200, please follow here.

For ScanNet200-SV, please follow here.

For ScanNet200-MV, please follow here.

ScanNet:

For ScanNet, please follow here.

For ScanNet-SV, please follow here.

For ScanNet-MV, please follow here.

SceneNN:

For SceneNN, please follow here.

For SceneNN-MV, please follow here.

3RScan:

For 3RScan, please follow here.

For 3RScan-MV, please follow here.