🎉 Welcome to the 3D Object Detection Hub

This site was born out of a Master’s thesis effort to centralize and compare every major 3D-OD method using multiple sensor modalities. While many prior surveys focus on one sensor or era of work, here you’ll find:

A unified, searchable, and maintainable catalog
covering camera-only, LiDAR-only, and multi-modal fusion approaches - all organized by sensor and representation.

If you use this site or data, please cite our publication:

Valverde, M., Moutinho, A., & Zacchi, J. V. (2025). A Survey of Deep Learning-Based 3D Object Detection Methods for Autonomous Driving Across Different Sensor Modalities. Sensors.

🔗 Navigate

• 📊 Datasets
  A concise table of the leading benchmarks (KITTI, nuScenes, Waymo, …) and their key properties.

• 🛠️ Models
  A comprehensive database of published 3D-OD methods.
  • Filter by sensor → representation
  • Search for a method by name
  • Sort by year, mAP, runtime, etc.

• 📚 References
  The full bibliography of every paper, library, and dataset used to create this work.

• 👤 About
  Background on the site, the underlying thesis, and contact details.

📖 Why This Hub?

Although 3D object detection has exploded over the last decade, existing resources often:

• ✔️ Cover only one modality (e.g. camera-only or LiDAR-only).
• ✔️ Become outdated as new methods appear.
• ✔️ Lack a unified taxonomy across sensors & representations.

This Hub aggregates:

1. Multi-sensor methods (monocular, stereo, multiview, point-cloud, voxels, fusion…).
2. Performance comparisons across KITTI, nuScenes & Waymo.
3. Runtime metrics to inform real-world feasibility.
4. Interactive filtering, searching, and sorting.

🚀 What You’ll Find

📚 Datasets & Benchmarks

• KITTI – The foundational benchmark (2 cameras + 1 LiDAR) with 15 000 annotated frames.
• nuScenes – Multi-modal data from 6 cameras, 1 LiDAR, 5 radars over 1 000 scenes.
• Waymo Open – High-density 5 × LiDAR + 5 × camera over 1 150 segments and 230 000+ frames.
  …and more (ApolloScape, Argoverse, Lyft Level 5, H3D).

🛠️ Methods & Models

• Monocular (Mono3D, SMOKE, DD3D, MonoFlex, …)
• Stereo & Multiview (LIGA-Stereo, Pseudo-BEV, …)
• LiDAR-only (PointPillars, PV-RCNN, SECOND, …)
• Multi-Modal Fusion (MVXNet, CLOCS, FUTR3D, …)

Each entry shows accuracy, inference time, paper link, and code availability.

🔍 How to Use

1. Browse the Datasets page to choose your benchmark.
2. Visit Models to filter by sensor & representation.
3. Search or sort by mAP, runtime, year, or code availability.
4. Click “Link” to read the original paper—code links when available.
5. Cite via the References section when you publish your own results! 📑

🙋‍♂️ About & Contact

For background on the thesis, site construction, and to reach out, visit 👉 About or drop me a line at ✉️ miguel.heitor.valverde@tecnico.ulisboa.pt.

Keep Exploring!
Whether you’re benchmarking a new sensor, prototyping a fusion network, or writing the next SOTA paper, the 3D Object Detection Hub is here to accelerate your research. 🔬🚗

🔮 Future Work

• 🔄 Continuously update with new methods & datasets.
• ⚙️ Introduce a Sensor Guide: strengths & trade-offs of cameras, LiDARs, radars.
• 📊 Expand to 2D OD and Depth Estimation surveys.
• 📝 Include references for a paper to be published, detailing how this study was carried out and its methodology.

📑 Citation

If you use this site or data, please cite our publication:

Valverde, M., Moutinho, A., & Zacchi, J. V. (2025). A Survey of Deep Learning-Based 3D Object Detection Methods for Autonomous Driving Across Different Sensor Modalities. Sensors.