Data Preparation

GPOcc is trained and evaluated on two datasets: OccScanNet and EmbodiedOcc-ScanNet.

Directory Structure

After preparation, the data/ directory should look like:

data/
├── occscannet/
│   ├── train_final.txt
│   ├── test_final.txt
│   ├── train_mini_final.txt
│   ├── test_mini_final.txt
│   ├── gathered_data -> /path/to/OccScanNet/gathered_data
│   └── posed_images  -> /path/to/OccScanNet/posed_images
└── scene_occ/
    ├── train_online.txt
    ├── test_online.txt
    ├── train_mini_online.txt
    ├── test_mini_online.txt
    ├── global_occ_package       -> /path/to/EmbodiedOcc-ScanNet/global_occ_package
    └── streme_occ_new_package   -> /path/to/EmbodiedOcc-ScanNet/streme_occ_new_package

OccScanNet

1. Download the dataset from hongxiaoy/OccScanNet.

2. Unzip the downloaded files.

3. Create symbolic links under data/occscannet/:

cd data/occscannet
ln -s /path/to/OccScanNet/gathered_data
ln -s /path/to/OccScanNet/posed_images
cd ../..

EmbodiedOcc-ScanNet

Used for the embodied (streaming) evaluation setting.

1. Download the dataset from YkiWu/EmbodiedOcc-ScanNet.

2. Unzip the downloaded files.

3. Create symbolic links under data/scene_occ/:

cd data/scene_occ
ln -s /path/to/EmbodiedOcc-ScanNet/global_occ_package
ln -s /path/to/EmbodiedOcc-ScanNet/streme_occ_new_package
cd ../..

Mini Sets

The split files *_mini_*.txt provide a small subset for quick validation runs. Set data_tg='mini' in the dataset config to use them.