---
license: cc-by-nc-4.0
task_categories:
- object-detection
---

# Boxer Dataset

This repository contains sample data for **Boxer: Robust Lifting of Open-World 2D Bounding Boxes to 3D**.

[Project Page](https://facebookresearch.github.io/boxer) | [Paper](https://huggingface.co/papers/2604.05212) | [GitHub](https://github.com/facebookresearch/boxer)

## Introduction
Boxer is an algorithm designed to estimate static 3D bounding boxes (3DBBs) from 2D open-vocabulary object detections, posed images, and optional depth. This repository hosts sample data sequences used to demonstrate Boxer's ability to lift 2D detections into 3D world space.

## Data Sources
The dataset provides sample sequences from several sources used in the paper:
* **Project Aria**: Sequences from Gen 1 & 2 (e.g., `hohen_gen1`, `nym10_gen1`, `cook0_gen2`).
* **CA-1M**: A sample validation sequence.
* **SUN-RGBD**: A subset of sample images.

## Usage

### Download Data
To download the sample data to your local machine, you can use the utility scripts provided in the [official GitHub repository](https://github.com/facebookresearch/boxer):

```bash
# Download sample Aria sequences
bash scripts/download_aria_data.sh

# Download CA-1M sample
python scripts/download_ca1m_sample.py

# Download SUN-RGBD sample
python scripts/download_omni3d_sample.py
```

### Run Inference
Once the data and checkpoints are downloaded, you can run BoxerNet on a sequence (e.g., `nym10_gen1`):

```bash
python run_boxer.py --input nym10_gen1 --max_n=90 --track
```

## Citation
```bibtex
@article{boxer2026,
      title={Boxer: Robust Lifting of Open-World 2D Bounding Boxes to 3D},
      author={Daniel DeTone and Tianwei Shen and Fan Zhang and Lingni Ma and Julian Straub and Richard Newcombe and Jakob Engel},
      year={2026},
}
```