Freeform

🚧

Coming Soon

Freeform dataset support is currently in development. This page describes the upcoming capability and its intended use cases. Check back for updates on availability and documentation.

Freeform datasets in Datature Vi enable you to define custom annotation schemas tailored to your specific use case. Unlike Phrase Grounding or Visual Question Answering which have predefined structures, freeform datasets provide maximum flexibility for specialized computer vision applications.

Think of freeform as a blank canvas where you can design your own annotation format. This is ideal for novel research, experimental projects, or industry-specific applications that don't fit into traditional vision task categories.

---

What is Freeform?

Freeform datasets allow you to create custom annotation formats that match your exact requirements. Rather than conforming to predefined structures like bounding boxes with labels or question-answer pairs, you can define:

• Custom data structures — Design annotation formats specific to your domain
• Flexible schemas — Mix and match different annotation types
• Novel vision tasks — Experiment with new computer vision paradigms
• Specialized workflows — Build annotation pipelines for unique use cases

Key characteristics

Maximum flexibility

• Define your own annotation structure
• No predefined format constraints
• Adaptable to changing requirements

Domain-specific

• Tailored to industry-specific needs
• Support for specialized metadata
• Custom validation rules

Research-friendly

• Ideal for experimental projects
• Support for novel vision tasks
• Flexible data iteration

---

When to use Freeform

Choose freeform datasets when your use case requires custom annotation formats that don't fit Phrase Grounding or VQA paradigms.

Ideal scenarios

Research projects

• Exploring novel computer vision tasks
• Developing new annotation paradigms
• Testing experimental model architectures

Specialized industries

• Medical imaging with custom diagnostic annotations
• Scientific imaging with domain-specific labels
• Industrial applications with unique requirements

Hybrid requirements

• Combining multiple annotation types
• Custom spatial annotations beyond bounding boxes
• Complex multi-modal annotations

Evolving use cases

• Prototyping new annotation workflows
• Iterating on annotation schemas
• Adapting to changing project requirements

---

Freeform vs. other dataset types

Understanding when to use freeform versus other dataset types:

Choose Freeform when:

• Standard formats don't meet your needs
• You're conducting research on novel vision tasks
• Your industry requires specialized annotations
• You need to combine multiple annotation types

Choose Phrase Grounding when:

• You need object detection with flexible descriptions
• You want spatial localization of objects
• Standard bounding box format works for your use case

Choose VQA when:

• You need conversational image understanding
• You want question-answer based annotations
• Text responses are sufficient

---

Common use cases

Research and experimentation

Novel vision tasks

• Developing new computer vision paradigms
• Testing experimental annotation strategies
• Exploring multi-modal learning approaches

Academic research

• Publishing novel datasets
• Creating reproducible benchmarks
• Advancing computer vision research

Specialized domains

Medical imaging

• Custom diagnostic annotations
• Specialized measurement formats
• Domain-specific metadata

Scientific research

• Microscopy image analysis
• Astronomical data annotation
• Custom scientific measurements

Industrial applications

• Manufacturing-specific defect classification
• Custom quality metrics
• Industry-specific spatial annotations

Advanced applications

Multi-task learning

• Combining detection, segmentation, and classification
• Hierarchical annotations
• Complex annotation relationships

Custom workflows

• Specialized annotation pipelines
• Domain-specific validation rules
• Custom data preprocessing

---

Getting started

🚧

Development in progress

Freeform dataset creation and annotation capabilities are currently being developed. When available, you'll be able to:

• Create freeform datasets via the Vi dashboard
• Define custom annotation schemas
• Upload freeform annotations via Vi SDK
• Download freeform annotations in Vi JSONL format

Contact us for updates on availability or to discuss your custom annotation requirements.

Future capabilities

When freeform support launches, you'll be able to:

1. Create freeform datasets — Define your custom annotation schema
2. Upload annotations — Import data in Vi JSONL format
3. Download annotations — Export for external processing
4. Train models — Fine-tune VLMs on custom data

---

Best practices

When freeform datasets become available, follow these guidelines:

Schema design

Keep it simple

• Start with the minimal required structure
• Add complexity only when necessary
• Document your schema thoroughly

Plan for iteration

• Design schemas that can evolve
• Version your annotation format
• Keep backward compatibility in mind

Data quality

Validate consistently

• Define clear validation rules
• Implement automated checks
• Maintain annotation quality standards

Document extensively

• Create annotation guidelines
• Document schema changes
• Provide examples for annotators

Integration

Consider downstream tasks

• Design with model training in mind
• Think about evaluation metrics
• Plan for inference pipeline integration

Maintain compatibility

• Use standard data types where possible
• Consider export format needs
• Plan for data migration

---

Alternative options

While freeform support is in development, consider these alternatives:

Use existing dataset types

If your use case can be adapted:

• Phrase Grounding for object localization needs
• VQA for question-based annotations

External processing

For immediate custom annotation needs:

• Annotate data externally
• Process and validate offline
• Import when freeform support launches

Contact Datature

For specialized requirements:

• Contact our team to discuss custom solutions
• Share your use case and requirements
• Get updates on freeform availability

---

Learn more

• Phrase grounding — Object localization with natural language
• Visual question answering — Question-answering for images
• Create a dataset — Dataset creation fundamentals
• Concepts overview — Core VLM concepts
• Contact us — Get in touch about custom requirements