View Dataset Insights

Dataset insights provide comprehensive analytics about your dataset composition, helping you understand annotation distribution, identify quality issues, and ensure your data is ready for training.

💡

Part of dataset workflow

Create dataset → Upload data → Annotate → View insights (you are here) → Train model

What insights are available?

Asset statistics

• Total number of assets in the dataset
• File format distribution
• File size analytics
• Upload timeline

Annotation metrics

• Total annotation count
• Phrase grounding pairs (if applicable)
• VQA question-answer pairs (if applicable)
• Annotations per asset

Distribution analysis

• Class balance across annotations
• Annotation type distribution

Quality indicators

• Assets without annotations

Why dataset insights matter

How to access insights

Via web interface

1. Navigate to your dataset
2. Click the Insights or Analytics tab
3. Review the displayed metrics and visualizations
4. Use filters to focus on specific aspects

Via Vi SDK

import vi

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

# Get dataset information
dataset = client.datasets.get("your-dataset-id")
print(f"Total assets: {dataset.asset_count}")
print(f"Annotations: {dataset.annotation_count}")

Learn more about Vi SDK →

Best practices

Common scenarios

Insufficient annotations

Problem: Dataset has too few annotations for training

Action:

• Add more annotations manually
• Upload additional annotations if available
• Use AI-assisted tools to speed up labeling

Class imbalance

Problem: Some classes heavily overrepresented

Action:

• Collect more data for underrepresented classes
• Remove excess samples from overrepresented classes
• Adjust dataset split strategy

Missing annotations

Problem: Many assets without annotations

Action:

• Complete annotation workflow
• Upload missing annotations if they exist
• Remove unannotated assets via bulk actions

Next steps

Related resources

• Manage datasets — Overview of dataset management operations
• Create a dataset — Set up new datasets with proper configuration
• Annotate data — Create quality annotations for training
• Upload data — Add more images and labels
• Manage assets — Delete, organize, and clean up images
• Download data — Export dataset for backup or external processing
• Configure your dataset — Adjust training splits and settings
• Train a model — Use your dataset for VLM training
• Resource usage — Monitor Data Rows and storage consumption
• Vi SDK Datasets API — Manage datasets programmatically