README.md

# 3DIEBench-OOD

Out-of-distribution (OOD) test set for [3DIEBench](https://github.com/facebookresearch/SIE/tree/main/data), designed to evaluate equivariance generalization to unseen rotation ranges.

## Dataset Description

3DIEBench-OOD contains rendered views of ShapeNet objects with rotations sampled from ranges **not seen during training**:

| | Training (3DIEBench) | OOD Test (This Dataset) |
|---|---|---|
| **Rotation Range** | [-π/2, π/2] | [-π, -π/2] ∪ [π/2, π] |

This allows testing whether models can extrapolate equivariant predictions to OOD rotation angles.

For generation details, see [`data/3DIEBench/data_generate_ood.py`](https://github.com/hafezgh/seq-jepa/blob/main/seq-jepa/data/3DIEBench/data_generate_ood.py) in the seq-JEPA repository.

### Structure

| File | Description |
|------|-------------|
| `{synset}/{obj}/image_{i}.jpg` | Rendered image (256×256) |
| `{synset}/{obj}/latent_{i}.npy` | 7D latent vector: [yaw, pitch, roll, floor_hue, spot_θ, spot_φ, spot_hue] |

## Usage

import numpy as np
from PIL import Image

img = Image.open('04330267/8ff18f81de484792f0b94599b4efe81/image_2.jpg')
latent = np.load('04330267/8ff18f81de484792f0b94599b4efe81/latent_2.npy')
# latent = [yaw, pitch, roll, floor_hue, spot_theta, spot_phi, spot_hue]## Related Resources

- **seq-JEPA Code**: [GitHub](https://github.com/hafezgh/seq-jepa)
- **Project Page**: [hafezgh.github.io/seq-jepa](https://hafezgh.github.io/seq-jepa/)
- **Original 3DIEBench**: [SIE Repository](https://github.com/facebookresearch/SIE/tree/main/data)

## Citation

If you use this dataset, please cite:

@inproceedings{ghaemi2025seqjepa,
  title={seq-{JEPA}: Autoregressive Predictive Learning of Invariant-Equivariant World Models},
  author={Ghaemi, Hafez and Muller, Eilif Benjamin and Bakhtiari, Shahab},
  booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
  year={2025},
  url={https://openreview.net/forum?id=GKt3VRaCU1}
}

@inproceedings{garrido2024sie,
  title={Self-supervised learning of split invariant equivariant representations},
  author={Garrido, Quentin and Balestriero, Randall and Najman, Laurent and LeCun, Yann},
  booktitle={International Conference on Machine Learning},
  year={2024}
}