MODEST

Multi-Optics DOF Stereo Dataset

A large-scale computational photography dataset for depth-of-field rendering, stereo vision, and optical scene understanding.

Project Website • Paper

---

Overview

MODEST is the first high-resolution stereo DSLR dataset that systematically captures variations in focal length and aperture using professional camera optics. The dataset addresses key limitations of existing depth-of-field datasets by combining ultra-high-resolution imagery, stereo geometry, and controlled optical parameter variations within complex real-world scenes.

The dataset contains 20,000 stereo images captured across 10 diverse scenes at 5472 × 3648 (20MP) resolution. Each scene is photographed using 10 focal lengths, 5 aperture settings per focal length, and multiple viewpoints, enabling comprehensive analysis of depth-of-field, defocus blur, optical aberrations, focal-length-dependent appearance changes, and multi-view scene understanding. Additionally, MODEST includes dedicated high-resolution checkerboard calibration image sets for every focal length, allowing recalibration using future calibration algorithms and facilitating camera geometry research.

The scenes include challenging visual phenomena such as reflective and semi-transparent surfaces, optical illusions, ambiguous depth relationships, fine structures, strong brightness transitions, and complex lighting conditions, making the dataset suitable for evaluating modern computer vision and computational photography methods under realistic conditions.

Highlights

• 20,000 ultra-high-resolution (20MP) stereo images
• 10 real-world scenes captured under controlled optical variations
• 10 focal lengths × 5 aperture settings per scene
• Synchronized stereo DSLR capture
• Dedicated high-resolution calibration image sets for every focal length
• Real-world depth-of-field, defocus blur, and optical aberrations
• Challenging reflective, transparent, and fine-detail scene content
• Suitable for computational photography, stereo vision, camera calibration, image restoration, and 3D scene understanding

Research Applications

MODEST supports research in:

• Depth-of-field rendering
• Defocus deblurring
• Stereo matching
• Multi-view learning
• Computational photography
• Camera calibration
• Optical scene understanding
• Image restoration
• 3D reconstruction

Contributions

1. Dataset Contribution We introduce MODEST, the first high-resolution stereo DSLR dataset with systematic focal length and aperture variations. The dataset combines optical realism, stereo geometry, and controlled camera parameter changes at a scale not previously available in public depth-of-field datasets.

2. Benchmarking and Analysis We provide a systematic evaluation of state-of-the-art depth-of-field rendering and defocus deblurring methods across varying optical configurations. Our analysis highlights challenges posed by reflective surfaces, semi-transparent materials, fine details, strong light sources, depth-varying blur, cat-eye vignetting effects, and non-uniform bokeh patterns.

3. Open Research Resources We release the complete MODEST dataset, calibration images and supporting materials to facilitate future research in computational photography and optical scene understanding for non-commercial academic use.

Dataset Statistics

Property	Value
Total Images	20,000
Resolution	5472 × 3648
Indoor Scenes	10
Camera	Canon 6D DSLR
Focal Lengths	10
Aperture Settings	5
Depth Range	0.5m – 10m
Optical Configurations	50
Dataset Size	~142 GB

---

Dataset Structure

MODEST/
├── Global_calibration_set/
│   ├── EOS6D_A_Left/
│   │   └── fl_<focal_length>/
│   │       ├── calibration/
│   │       └── inference/
│   │
│   └── EOS6D_B_Right/
│       └── fl_<focal_length>/
│           ├── calibration/
│           └── inference/
│   
│   
│
├── Scene<id>/
│   ├── EOS6D_A_<Left|Right>/
│   │   └── fl_<focal_length>/
│   │       ├── calibration/
│   │       └── inference/
│   │           └── F<aperture>/
│   │           
│   │
│   └── EOS6D_B_<Left|Right>/
│       └── fl_<focal_length>/
│           ├── calibration/
│           └── inference/
│               └── F<aperture>/
│               
│
└── ...

Notes

<id> ∈ {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}

<focal_length> ∈ {28mm, 32mm, 36mm, 40mm, 45mm, 50mm, 55mm, 60mm, 65mm, 70mm}

<aperture> ∈ {F2.8, F5.0, F9.0, F16.0, F22.0}

Capture System

Stereo acquisition system used for capturing the MODEST dataset.

MODEST was captured using a synchronized stereo rig consisting of two Canon 6D DSLR cameras mounted on a custom stereo platform. The setup enables controlled acquisition across multiple focal lengths and aperture settings while preserving scene geometry and viewpoint consistency.

Project Resources

Project Website

https://modest-dataset.netlify.app/

Paper

https://arxiv.org/abs/2511.20853

---

Citation

If you use MODEST in your research, please cite:

@misc{trivedi2026modestmultiopticsdepthoffieldstereo,
      title={MODEST: Multi-Optics Depth-of-Field Stereo Dataset}, 
      author={Nisarg K. Trivedi and Vinayak A. Belludi and Li-Yun Wang},
      year={2026},
      eprint={2511.20853},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.20853}, 
}

---

License

This dataset is released under the CC BY 4.0 License.

Users are free to share and adapt the dataset provided appropriate attribution is given to the authors.

---

Acknowledgements

MODEST was created to support open research in computational photography, stereo vision, and optical image formation.

We hope it serves as a valuable benchmark for future work in vision and imaging research.