Download Full Dataset

Downloading the full dataset exports both your assets (images, videos) and their associated annotations in a structured format. This is essential for creating backups, local development, external training workflows, or migrating data to other platforms.

💾

What's included in a full dataset export

• All asset files — Original images or videos in your dataset
• Annotation files — Complete annotation data in structured format
• Test split — Optional automatic train/test split organization
• Normalized data — Assets and annotations organized in training-ready structure

💻

Programmatic access with Vi SDK

You can also download datasets programmatically using the Vi SDK:

• Download datasets — Use client.get_dataset() to download programmatically
• Load and iterate — Use ViDataset to iterate through asset and annotation pairs
• Visualize annotations — Built-in visualization tools to render annotations over assets
• Automate workflows — Integrate dataset downloads into your pipelines

Learn more about Vi SDK →

Navigate to annotation export

1. In your organization, navigate to the Explorer section from the sidebar

2. Select the dataset you want to download from the dataset list

3. Click the Annotations tab in the Dataset Explorer header

4. The Annotations page displays the Export Annotations section with an Export button

📘

Dataset Explorer - Annotations tab

The Export Annotations section allows you to export your dataset in various formats.

Export your full dataset

Open the export dialog

1. Click the Export button in the Export Annotations section

2. The Annotation Export dialog opens with configuration options

Configure export settings

Export Format

1. In the Export Format dropdown, select Full Dataset

   This option exports both your assets and annotations together in a structured format.

📘

Full Dataset export configuration

Configure the export format, test split ratio, and normalization options before exporting.

Test Split Ratio (Optional)

The test split ratio automatically divides your dataset into training and testing subsets:

1. Check the Enabled checkbox to activate test splitting

2. Enter a decimal value between 0 and 1 (e.g., 0.2 for 20% test, 80% train)

   • 0.0 — No split; all data saved in the dump/ folder
   • 0.1 — 10% for testing, 90% for training
   • 0.2 — 20% of data for testing, 80% for training
   • 0.3 — 30% for testing, 70% for training

3. Leave unchecked (or set to 0) if you want the entire dataset without splitting

📊

Test split best practices

• No split (0.0) — All data saved in dump/ folder; useful for backups or custom splitting
• Standard split — 0.2 (80/20) is commonly used for most datasets
• Large datasets — 0.1 (90/10) works well when you have thousands of images
• Small datasets — 0.2 or 0.3 ensures sufficient test data for validation
• Stratified splits — Maintains class distribution across train/test sets

Normalization

Select the Normalized option to organize your export in a training-ready structure:

• Normalized — Assets and annotations organized by class/category for easy training
• Creates structured folders compatible with common ML frameworks

Export preview

The dialog displays a summary of what will be exported:

You are about to export [N] assets with [M] annotations.

This helps you verify the export scope before downloading.

Download the dataset

1. Review your configuration settings:

   • Export Format: Full Dataset
   • Test Split Ratio: Your chosen value or disabled
   • Normalization: Normalized

2. Click the Export button to start the download

3. The export job begins processing and will appear in the Annotation Job History section below

4. Once completed, the download will begin automatically or a download link will be provided

Export structure

The full dataset export is organized in a structured format for easy use:

Directory structure

With train/test split enabled (split ratio > 0):

dataset-name/
├── train/
│   ├── images/
│   │   ├── image1.jpg
│   │   ├── image2.jpg
│   │   └── ...
│   └── annotations/
│       ├── image1.json
│       ├── image2.json
│       └── ...
├── test/
│   ├── images/
│   │   ├── test1.jpg
│   │   └── ...
│   └── annotations/
│       ├── test1.json
│       └── ...
└── metadata.json

Without train/test split (split ratio = 0 or disabled):

dataset-name/
├── dump/
│   ├── assets/
│   │   ├── image1.jpg
│   │   ├── image2.jpg
│   │   └── ...
│   └── annotations/
│       └── annotations.jsonl
└── metadata.json

Annotation format

Each annotation file contains:

• Image metadata — Dimensions, filename, source path
• Annotation objects — Bounding boxes, labels, segmentation data
• Class information — Category IDs and names
• Attributes — Custom attributes if configured

Working with downloaded datasets

Once you've downloaded your dataset, you can work with it programmatically using the Vi SDK.

import vi

# Initialize client
client = vi.Client(
    secret_key="your-secret-key",
    organization_id="your-organization-id"
)

# Download dataset
result = client.get_dataset(
    dataset_id="your-dataset-id",
    save_dir="./data"
)

print(result.summary())
print(f"Downloaded to: {result.save_dir}")

from vi.dataset.loaders import ViDataset

# Load the downloaded dataset
dataset = ViDataset("./data/your-dataset-id")

# Get dataset info
info = dataset.info()
print(f"Total assets: {info.total_assets}")
print(f"Total annotations: {info.total_annotations}")

# Iterate through training data
for asset, annotations in dataset.training.iter_pairs():
    print(f"\n🖼️  {asset.filename}")
    print(f"   Size: {asset.width}x{asset.height}")

    for ann in annotations:
        if hasattr(ann.contents, 'caption'):
            # Phrase grounding annotation
            print(f"   📝 Caption: {ann.contents.caption}")
            print(f"      Bounding boxes: {len(ann.contents.grounded_phrases)}")
        elif hasattr(ann.contents, 'interactions'):
            # VQA annotation
            print(f"   💬 Q&A pairs: {len(ann.contents.interactions)}")

from vi.dataset.loaders import ViDataset

# Load dataset
dataset = ViDataset("./data/your-dataset-id")

# Get first asset and its annotations
asset, annotations = next(dataset.training.iter_pairs())

# Visualize annotations on the asset
asset.visualize(
    annotations=annotations,
    save_path="output.jpg",  # Optional: save to file
    show=True  # Display the visualization
)

Complete Vi SDK documentation →

Common use cases

Best practices

Monitor export progress

The Annotation Job History section tracks all export operations:

• Job type — Upload or Export
• Initiated by — User who started the export
• File count — Number of files included
• Status — In Progress, Finished, or Failed
• Timestamp — When the job completed

You can:

• View job details — Click on a job to see more information
• Track progress — Monitor active exports
• Redownload — Access completed exports if available

Troubleshooting

Next steps