search
Ctrlk
Observ

Release 2025/6/10

hashtag
June Release!

We’re excited to introduce several key enhancements and experimental features in this month’s DataVerse update. These improvements aim to streamline annotation workflows, improve performance, and open the door to powerful AI-assisted data generation and prediction.

hashtag
New Features

hashtag
VQA Import from Existing Projects

You can now import annotated VQA data from other projects directly into a new VQA project. This enables smoother cross-project dataset reuse and collaboration.

hashtag
Annotation Experience Improvements

  • Undo Support: Use Ctrl+Z to undo the last annotation action.

  • UI/UX Enhancements: Improved interaction behaviors during annotation.

  • Task List Filters: Easily search, filter, and jump across pages in task lists for faster access to your labeling work.

hashtag
Dark Mode UI

A new night mode interface is now available, making long annotation sessions easier on the eyes in low-light conditions.

hashtag
Data Import Performance Optimization

The backend system for data import has been significantly optimized to reduce processing and wait time, allowing for faster ingestion of large datasets.

hashtag
🧪 Beta Features (Available Upon Request)

If you’d like to enable the following beta features, please contact us for early access:

hashtag
AI-Powered Data Generation

Generate synthetic data across multiple modalities:

  • Text-to-Video, Image-to-Video, and more Ideal for rare or edge-case scenarios where data diversity is limited.

2. VQA Prediction & Analysis

Run Visual Language Model (VLM) predictions on your VQA tasks. Automatically generate answers and analyze model performance across different question types—turning annotation into actionable insight.

hashtag
Coming Soon to DataVerse

Data Deduplication:

To further improve data quality and streamline dataset preparation, we’re introducing Data Deduplication.

This feature will allow you to automatically identify and remove near-duplicate images based on a configurable similarity threshold. It’s especially useful when collecting large volumes of real-world data, helping you reduce redundancy and maintain a more diverse, efficient dataset.

Stay tuned—this feature is coming soon to DataVerse!

hashtag
Support

For technical support, please contact our support teamarrow-up-right during business hours.

PreviousRelease 2025/8/25NextRelease 2025/4/10

Last updated