# Data Card for Break4Models

## Dataset Description

- **Dataset Name**: Break4Models
- **Dataset Type**: Physics simulation data
- **Language**: English
- **License**: PDDL (Public Domain Dedication and License)
- **Created by**: [FractureRB](https://github.com/david-hahn/FractureRB)

### Dataset Sources

- **Repository**: [GitHub Repository](https://github.com/nikoloside/TEBP)
- **Paper**: [DeepFracture: A Generative Approach for Predicting Brittle Fractures with Neural Discrete Representation Learning](https://onlinelibrary.wiley.com/doi/abs/10.1111/cgf.70002)
- **Project Page**: [Project Page](https://nikoloside.graphics/deepfracture/)

## Dataset Overview

Break4Models is a small-scale 4-shape dataset for deepfracture paper. It contains simulation data of various 3D objects undergoing impact and breaking fracture pattern, designed for training and evaluating neural networks that can predict object fracture patterns for specific one target shape.

### Dataset Structure

The dataset is organized by object categories:
- `_out_base/` - Base object simulations (277 samples)
- `_out_pot/` - Pot object simulations (433 samples)
- `_out_squirrel/` - Squirrel object simulations (529 samples)
- `_out_bunny/` - Bunny object simulations (500 samples)

Each category contains:
- `meta.json` - Metadata with simulation statistics
- `obj/` - 3D mesh files (.obj format) for each simulation step
- `info/` - Collision information files
- `impact/` - Detailed impact data in JSON format
- `initial_cond/` - Initial conditions for simulations
- `nii/` - NIfTI format data (if applicable)
- `gif/` - Animation files showing breaking process

## Data Collection

### Source Data

The dataset was created by [FractureRB](https://github.com/david-hahn/FractureRB).

### Annotations

- **Impact Data**: JSON format containing collision information
  - `collElems`: Collision element IDs
  - `collPoints`: Collision point coordinates [x, y, z]
  - `collDirections`: Collision direction vectors
  - `collVels`: Collision velocities
  - `collImpulse`: Collision impulse values
  - `collForce`: Collision force magnitudes

- **Info Files**: Simple format `directory/,simulation-index-number`

### Personal and Sensitive Information

This dataset contains no personal or sensitive information as it consists entirely of synthetic physics simulation data.

## Dataset Statistics

### Dataset Size

- **Total Objects**: 4 target shapes (base, pot, squirrel, bunny)
- **Total Simulations**: Varies by category
  - Base: 277 simulations
  - Pot: 433 simulations
  - Squirrel: 529 simulations
  - Bunny: 500 simulations

### Data Fields

- `target_shape`: The target object shape name
- `max_gssdf_val`: Maximum GS-SDF (Geometrically-Segmented Signed Distance Field) value
- `avg_max_gssdf_val`: Average maximum GS-SDF value
- `min_gssdf_val`: Minimum GS-SDF value
- `avg_min_gssdf_val`: Average minimum GS-SDF value
- `start_time`: Simulation start time
- `stop_time`: Simulation end time
- `total_valid`: Total number of valid simulation times

## Uses

### Direct Use

This dataset is designed for:
- Training neural networks for physics-based collision-conditional brittle fracture
- Evaluating fracture prediction models for deepfracture paper
- Research in computational physics and computer graphics
- Demo for real-time brittle fracture simulation systems

### Out-of-Scope Use

- Medical applications
- Safety-critical systems
- High-precision engineering simulations without additional validation

## Bias, Risks, and Limitations

### Bias

The dataset is generated from synthetic simulations and may not perfectly represent real-world physics. The objects and materials are idealized and may not capture all real-world complexities.

### Risks

- Models trained on this data may not generalize to real-world scenarios
- Performance may degrade with objects significantly different from training data
- No guarantees for physical accuracy in safety-critical applications

### Limitations

- Limited to the specific object categories in the dataset
- Synthetic data may not capture all real-world physics phenomena
- Requires significant computational resources for processing

## Training and Evaluation

### Training Data

The dataset is split into training and test sets:
- **Training set**: 200 samples
- **Test set**: 50 samples

This split is used for model development and evaluation.

### Evaluation Data

Evaluation metrics include:
- Geometric accuracy
- Physics consistency
- Computational efficiency

## Citation

```bibtex
@article{huang2025deepfracture,
  author = {Huang, Yuhang and Kanai, Takashi},
  title = {DeepFracture: A Generative Approach for Predicting Brittle Fractures with Neural Discrete Representation Learning},
  journal = {Computer Graphics Forum},
  pages = {e70002},
  year = {2025},
  keywords = {animation, brittle fracture, neural networks, physically based animation},
  doi = {https://doi.org/10.1111/cgf.70002},
  url = {https://onlinelibrary.wiley.com/doi/abs/10.1111/cgf.70002},
  eprint = {https://onlinelibrary.wiley.com/doi/pdf/10.1111/cgf.70002}
}
```

## Data Card Authors

Huang Niko(nikoloside), Fang Chaowei(fangsunjian)

## Data Card Contact

https://github.com/nikoloside/TEBP