ScenesPlat 7K
High-fidelity indoor 3D scenes from 8 data sources, ready for visual reconstruction, realistic rendering or vision-language training.
7,916 scenes, 11.27 billion 3DGS points, `.npy`, `.json` files and metadata in `.csv`
CC-BY-SA-4.0
Description
Scenesplat-7k is a vast open-source dataset composed of 7,916 3D scenes reconstructed from renowned datasets such as ScanNet, Matterport3D or Replica. It brings together over 11 billion Gaussian Splatting Points (3DGS), organized for advanced tasks in computer vision, virtual reality, and realistic 3D rendering. Scenes are provided with pose files, camera intrinsics, and annotations to facilitate training and test pipelines.
What is this dataset for?
- Train NerF or 3D Gaussian Splatting models for photorealistic rendering.
- Evaluate indoor 3D reconstruction algorithms based on RGB-D images.
- Test combined 3D vision/language architectures (e.g. joint vision-language training).
Can it be enriched or improved?
Yes, users can enrich this dataset by adding their own scenes captured in the same format. Additionally, additional annotations (language, semantics, objects) can be added in JSON or `.npy` files. Since the data is open, it is also possible to filter the scenes according to quality criteria (PSNR, SSIM, etc.) using the statistics provided.
🔎 In summary
🧠 Recommended for
- 3D vision researchers
- VR/AR developers
- NerF projects
🔧 Compatible tools
- Instant-NGP
- Gaussian Splatting
- PyTorch3D
- Blender
- Hugging Face Scripts
💡 Tip
Use PSNR/SSIM metrics to keep only high-fidelity scenes during your experiments.
Frequently Asked Questions
Does this dataset contain outdoor scenes?
No, all the scenes are indoor, from datasets like Matterport3D or ScanNet.
Can scenes be used in Blender or Unity?
Yes, after converting `.npy` files to standard formats like `.ply` or `.obj`, they can be used in these engines.
Can we filter scenes according to their visual quality?
Yes, `.csv` files give PSNR scores, SSIM, and other metrics to sort scenes according to your criteria.
