Download ablation dataset

In addition to our full OpenMaterial dataset, we have designed a controlled dataset. To verify the validity of the method for different materials, shapes and light conditions, you can start with our ablation dataset:

python download.py --token {your-token} --type ablation

Download main dataset

You can directly download main dataset here. if you download directly without our script, please run:

tar -xvf {tar_path} -C {LOCAL_DIR}

Because the dataset is split across multiple files for storage, we strongly recommend that you use our script to download different parts of the dataset.(Please refer to the file README.md for more information.)

The dataset is classified and named by material type. We use hugginface to store the dataset. Before using our script, you need to log in to your Huggingface account and get your access token on the "Settings" page. Then, if you need the complete dataset, you can run the following command:

python download.py --token {your-token} --type all

Or if you need to download a subset of a certain material type separately, such as "conductor", you can run the following command:

python download.py --token {your-token} --type conductor

After download our dataset correctly, the following file structure will be obtained:

            datasets
            ├── groundtruth
            │   ├── 5c4ae9c4a3cb47a4b6273eb2839a7b8c
            │   │   └── clean_5c4ae9c4a3cb47a4b6273eb2839a7b8c.ply
            │   ├── 5c0514eae1f94f22bc5475fe0970cd28
            │   │   └── clean_5c0514eae1f94f22bc5475fe0970cd28.ply
            │   └── ... 
            ├── openmaterial
            │   ├── 5c4ae9c4a3cb47a4b6273eb2839a7b8c
            │   │   ├── train
            │   │   │   ├── images
            │   │   │   ├── mask
            │   │   ├── test
            │   │   ├── transforms_train.json
            │   │   └── transforms_test.json
            │   ├── 5c0514eae1f94f22bc5475fe0970cd28
            │   └── ... 
        

Optional depth data

Researchers may need to use depth maps for supervision or evaluation. We provide high-precision depth metadata instead of depth maps as optional downloadable items. The depth data is named "depth-xxxxxx.tar". You can directly download them on huggingface. Also, you can use our script with the depth parameter.

Or if you need to download depth data, please run following command:

python download.py --token {your-token} --type all --depth

Use depth data

an example for using depth data (Here are the real depth values, not normalised):

            with h5py.File(filename, 'r') as hdf:
                dataset = hdf['depth']
                depth = dataset[:] # size: (1200, 1600)