Data Types

2 minute read

The W&B Python SDK includes data types for logging various forms of media and structured data.

Overview

Data Types in W&B are classes that wrap media and structured data for logging to runs. They include visualization components in the W&B UI and handle data serialization, storage, and retrieval.

Available Data Types

Data Type	Description
`Image`	Log images with support for masks, bounding boxes, and segmentation.
`Video`	Track video data for model outputs or dataset samples.
`Audio`	Log audio samples for audio processing tasks.
`Table`	Create tables that can contain mixed media types.
`Plotly`	Log Plotly charts for data visualization.
`Html`	Embed custom HTML content.
`Object3D`	Visualize 3D point clouds and meshes.
`Molecule`	Log molecular structures for computational chemistry.
`Box3D`	Track 3D bounding boxes for 3D object detection.

Getting Started

Using Image Data Type to log an image:

import wandb
import matplotlib.pyplot as plt

# Generate an image for demonstration purposes
path_to_img = "/path/to/cafe.png"
im = plt.imread(path_to_img)

# Initialize a new run
with wandb.init(project="awesome-project") as run:

    # Log the image
    run.log({"img": [wandb.Image(im, caption="Cafe")]})

Using a Table Data Type to log a table with mixed text and labels:

import wandb

# Initialize a new run
with wandb.init(project="visualize-predictions", name="tables") as run:

    # Create tabular data, using a list of lists
    data = [["Cat", "1", "1"],["Dog", "0", "-1"]]
    run.log({"Table 1": wandb.Table(data=data, columns=["Text", "Predicted Label", "True Label"])})

    # Create tabular data, using `wandb.Table.add_data()` method
    table = wandb.Table(columns=["Text", "Predicted Label", "True Label"])
    table.add_data("Cat", "1", "1")
    table.add_data("Dog", "0", "-1")
    run.log({"Table 2": table})

Data Types

Overview

Available Data Types

Getting Started

Audio

box3d()

Html

Image

Molecule

Object3D

Plotly

Table

Video

Feedback