File size: 2,688 Bytes
e316634 5659295 e316634 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 |
---
license: apache-2.0
---
# SweepNet Dataset
This repository contains three datasets used in [SweepNet](https://arxiv.org/abs/2407.06305). One dataset comprised of 20,000 sweep surfaces for neural sweeper training and two datasets used in quantitative evaluations. All datasets are preprocessed.
## Neural Sweeper Dataset
We created 20,000 sweep surface samples to train the neural sweeper, please refer to the supplementary material for the training details.
We provided sweep surfaces with 3, 4 and 5 control points, structured as follows:
```
neuralSweeperData/
├── control_point_i/
│ ├── sweep_surface_index/
│ │ ├── parameterse.txt # sweep surface parameters
│ │ ├── bspline.ply #visualized sweeping axis
│ │ ├── sample_profile.obj #visualized profile
│ │ ├── result_sweep.ply # sweep surface
│ │ ├── manifold_points.npy # key points on the sweep surface
│ │ ├── sweep_occupancy_v1.npy # Occupancy field of the sweep surface
```
## GC-Object Dataset
We sampled 50 generalised cylinder featured objects from internet and prior works [OreX](https://arxiv.org/abs/2211.12886), [GCD](https://vcc.tech/research/2015/GCD#:~:text=Our%20decomposition%20algorithm%20progressively%20builds,on%20decomposition%20to%20global%20optimization.).
We provide processed 3D models here. Please consider cite us and the prior works if you find the dataset useful.
```
GC_objects/
├── model name/
│ ├── oracle.obj # Oracle 3D model (not the input)
│ ├── voxel_64_mc.off # 3D model reconstructed from input voxel
│ ├── skeletal_prior.ply # Model skeletons
│ ├── model_surface_point_cloud.ply # Surface point cloud for point cloud input modality
├── test_names.npz # List of all model names
├── voxel2pc.hdf5 # Model voxels and occupancy field used for training
├── ae_voxel_points_samples.hdf5 # Model voxels and occupancy field used *only* for testing
```
## Quadrupeds Dataset
We use quadrupeds dataset from [Tulsiani et al.](https://github.com/shubhtuls/volumetricPrimitives/issues/7) to benchmark SweepNet. We provide the processed data here, please cite us if you used our processed data.
```
quadrupeds/
├── model name/
│ ├── oracle.obj # Oracle 3D model (not the input)
│ ├── skeletal_prior.ply # Model skeletons
│ ├── model_surface_point_cloud.ply # Surface point cloud for point cloud input modality
├── test_names.npz # List of all model names
├── voxel2pc.hdf5 # Model voxels and occupancy field used for training
├── ae_voxel_points_samples.hdf5 # Model voxels and occupancy field used *only* for testing
``` |