Skip to content

Latest commit

 

History

History
58 lines (52 loc) · 2.4 KB

readme_data.md

File metadata and controls

58 lines (52 loc) · 2.4 KB
Raw

To download the dataset:

bash dowload_dataset.sh

This will download and unzip the dataset in the data folder and will unzip pre-trained models. The data directory has the following organization.

├── shapes
│   ├── all_ids.txt
│   ├── face_data_release.zip
│   ├── meshes.zip
│   ├── test_data.h5
│   ├── test_ids.txt
│   ├── train_data.h5
│   ├── train_ids.txt
│   ├── val_data.h5
│   └── val_ids.txt
└── spline
    ├── closed_splines.h5
    ├── open_splines.h5
    └── simple_less_thn_20.zip

  • meshes.zip: contains all the meshes used in the parsenet experiments. Note that these models are taken from ABC dataset. We pre-processed shapes to separate disconnected meshes into different meshes. For that reason you will notices that names of the shapes is of the format shapeid_index.json, where shapeid is the id of the model from ABC dataset and index is the index of the disconnected part.

  • train_data.h5, train_data.h5 and val_data.h5: contain points, normals, segment index and primitive type index for each shape. Please refer to src/dataset_segments.py on how to load these h5 files. Note that, for primitive types, there are possible 10 primitives, for example circle, sphere, plane, cone, cylinder, open spline, closed spline, revolution, extrusion and extra. revolution, extrusion and extra are treated as b-spline primitives because b-spline can also approximate these patches. Excluding shapes with these extra surface patches would have resulted in very small dataset. More specifically:

    1. [0, 6, 7, 9] indices correspond to closed b-spline.
    2. [2, 8] indices correspond to open b-spline.
    3. [1] index corresponds to plane.
    4. [3] corresponds to cone.
    5. [4] corresponds to cylinder.
    6. [5] corresponds to sphere.

  • face_data_release.zip: contains txt files for each shape in the above dataset. Specifically, it contains the segment id and primitive types for each shape.

  • train_ids.txt, val_ids.txt and test_ids.txt: contains shape ids for different splits. all_ids.txt contains list of ids for all shapes.

  • closed_splines.h5: contains points and control points for closed splines. Please refer to src/dataset.py for more details on how to load points, and splits. open_splines.h5 is for open splines.