|
|
--- |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- vision |
|
|
- point cloud |
|
|
- NYU Depth V2 |
|
|
- 3d |
|
|
- deep learning |
|
|
- computer vision |
|
|
- RAY-AUTRA-TECHNOLOGY |
|
|
language: en |
|
|
pretty_name: RAY-tech img_pointV2 |
|
|
datasets: |
|
|
- jagennath-hari/nyuv2 |
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
# img_pointV2 is available 🎉🎉🎉🥳🥳😀😀 |
|
|
|
|
|
This dataset is a collection of 3D point clouds generated from the `jagennath-hari/nyuv2dataset`. |
|
|
|
|
|
**img_pointV2** is the second version of the `RAY-AUTRA-TECHNOLOGY/img_pointV` dataset. It is a spatialized version of the *NYU Depth V2* dataset, transforming classic indoor images into high-fidelity 3D point clouds (`.ply` files). |
|
|
|
|
|
The main objective is to provide clean, ready-to-use 3D scenes for training 3D vision models, eliminating the need for users to manually handle RGB-D to point cloud conversion. |
|
|
|
|
|
--- |
|
|
|
|
|
### Dataset Highlights |
|
|
|
|
|
* **Point Clouds (.ply):** Complete 3D scenes featuring both geometry ($X, Y, Z$) and color ($R, G, B$). |
|
|
* **Metric Precision:** Every point is accurately positioned in meters, strictly following the real-world Kinect camera intrinsic parameters. |
|
|
* **Cleaned & Uniformed:** Clouds have been filtered to remove capture noise and voxelized with a 1 cm density (voxel size: $0.01$). |
|
|
* **Integrated Labels:** Metadata preserves all original semantic and instance segmentation information. |
|
|
|
|
|
--- |
|
|
|
|
|
### File Structure |
|
|
|
|
|
| File/Folder | Description | |
|
|
| :--- | :--- | |
|
|
| `data/` | Directory containing the `.ply` files. | |
|
|
| `metadata.arrow` | Central index linking IDs, filenames, and point counts (Train/Val/Test splits). | |
|
|
| `camera_params.json` | Optical parameters (intrinsics) used for the 3D reconstruction. | |
|
|
| `class_names.json` | Dictionary of semantic classes (e.g., chair, wall, table). | |
|
|
| `config.yaml` | Dataset configuration (license, format, normalization). | |
|
|
|
|
|
--- |
|
|
|
|
|
> **IMPORTANT:** These files are fully compatible with major 3D libraries such as **Open3D**, **PyTorch Geometric**, and **PointNet++**. |
|
|
|
|
|
RAY AUTRA TECHNOLOGY 2025 |
|
|
|
