---
license: apache-2.0
tags:
- vision
- point cloud
- imagenet
- 3d
- deep learning
- computer vision
language: en
pretty_name: RAY-tech img_pointV1
datasets:
- visual-layer/imagenet-1k-vl-enriched
---

![CLOUD_POINTS _dAtAsEt_](https://cdn-uploads.huggingface.co/production/uploads/66de3482fd7d68a29319ecd9/3E3G88NiZGaiZYnBvNbC1.png)


# img_pointV1

This dataset is a collection of **3D point clouds** generated from images in the **ImageNet-1k VL Enriched** dataset (`visual-layer/imagenet-1k-vl-enriched`).

Each 2D image is converted into a point cloud where the (X, Y) coordinates correspond to pixel locations, and the **Z coordinate (depth/elevation)** is derived from the pixel's **grayscale intensity**. The original image colors are retained as the colors of the points. The point clouds are stored in the **GLB** format.

---

## Dataset Structure

The dataset consists of two main parts:

1.  **`dataset_cloudV1.arrow`**: An **Apache Arrow** file containing the metadata for each point cloud.
2.  **`glb_files/`**: A directory containing the individual **GLB** files, each representing a 3D point cloud.

---

This process creates a 3D representation of the image, similar to a **heightmap**, where luminance dictates depth.


RAY AUTRA TECHNOLOGY 2025

---