swanlab.Object3D

GitHub Source Code

Parameter	Description
data	(Union[np.ndarray, str, Path]) Accepts point cloud file paths or NumPy arrays. The Object3D class will determine the received data type and perform corresponding conversions.
caption	(str) Label for the 3D object. Used to mark the 3D object when displayed in the experiment dashboard.

Introduction

Converts various types of point cloud data to be logged by swanlab.log().

Creating from Files/Dictionaries

Example File

data.swanlab.pts.json: Download from Google Drive

The file format is json, with the following content structure:

{
    "points": [
        [x1, y1, z1, r1, g1, b1],
        [x2, y2, z2, r2, g2, b2],
        ...
    ],
    // (Optional) Detection boxes for tasks such as point cloud detection, highlighting corresponding positions
    "boxes": [
        {
            "color": [r, g, b],
            "corners": [[x1,y1,z1], ..., [x8,y8,z8]],
            // (Optional) Label text for the detection box, displayed in the view
            "label": "class_name",
            // (Optional) Confidence score, displayed in the view
            "score": 0.95,
        },
        ...
    ]
}

Detailed Explanation of JSON File Parameters:

• points:

  • This is an array used to store 3D point cloud data.
  • Each element is an array of 6 numerical values [x, y, z, r, g, b], representing:
    • x, y, z: 3D coordinates of the point.
    • r, g, b: Color of the point, representing red, green, and blue channel values, typically in the range 0-255.

• boxes (Optional):

  • This is an array used to store 3D detection box data.
  • Each element is an object representing a detection box, containing the following fields:
    • color: Color of the detection box, an [r, g, b] array representing red, green, and blue channel values.
    • corners: Coordinates of the eight vertices of the detection box, an [[x1, y1, z1], ..., [x8, y8, z8]] array, where each element is a 3D coordinate [x, y, z].
    • label (Optional): Label text of the detection box, a string, used to display the category of the detection box in the view.
    • score (Optional): Confidence score of the detection box, a numerical value, typically in the range 0-1, representing the reliability of the detection box.

---

Using SwanLab to log 3D point cloud data from a json file:

Object3D

import swanlab

swanlab.init()

obj = swanlab.Object3D("data.swanlab.pts.json", caption="3d_point_cloud")
swanlab.log({"examples": obj})

Object3D.from_point_data

import swanlab

swanlab.init()

with open("data.swanlab.pts.json", "r") as f:
    cloud_point = json.load(f)

obj = swanlab.Object3D.from_point_data(
    points=cloud_point["points"],
    boxes=cloud_point["boxes"],
    caption="3d_point_cloud"
)

swanlab.log({"examples": obj})

Creating from NumPy Arrays

Creating from Coordinates

import numpy as np

# Example 1: Create point cloud from coordinates
points_xyz = np.array([
    [0, 0, 0],  # Point1: x=0, y=0, z=0
    [1, 1, 1],  # Point2: x=1, y=1, z=1
    [2, 0, 1]   # Point3: x=2, y=0, z=1
])

cloud_xyz = swanlab.Object3D(points_xyz, caption="Basic XYZ Points")
swanlab.log({"examples": cloud_xyz})

Creating from Coordinates and Categories

import numpy as np

# Example 2: Create point cloud with categories
points_xyzc = np.array([
    [0, 0, 0, 0],  # Point1: xyz + category 0
    [1, 1, 1, 1],  # Point2: xyz + category 1
    [2, 0, 1, 2]   # Point3: xyz + category 2
])

cloud_xyzc = swanlab.Object3D(points_xyzc, caption="Points with Categories")
swanlab.log({"examples": cloud_xyzc})

Creating from Coordinates and RGB

import numpy as np

# Example 3: Create point cloud with RGB colors
points_xyzrgb = np.array([
    [0, 0, 0, 255, 0, 0],    # Point1: xyz + red
    [1, 1, 1, 0, 255, 0],    # Point2: xyz + green
    [2, 0, 1, 0, 0, 255]     # Point3: xyz + blue
])

cloud_xyzrgb = swanlab.Object3D(points_xyzrgb, caption="Colored Points")
swanlab.log({"examples": cloud_xyzrgb})

Logging Multiple Point Clouds in a Single Step

import swanlab

...

cloud1 = swanlab.Object3D(points1, caption="cloud1")
cloud2 = swanlab.Object3D(points2, caption="cloud2")
cloud3 = swanlab.Object3D(points3, caption="cloud3")

...

swanlab.log({"examples": [cloud1, cloud2, cloud3, ...]})