Upload Data

Uploading data to your datasets is the foundation of training effective vision language models. Datature Vi provides flexible options for uploading both images (assets) and their corresponding annotations, whether you're working with a few dozen images or thousands.

This guide covers everything you need to know about getting your data into Datature Vi quickly and efficiently.

💡

Quick workflow

Create dataset → Upload data (you are here) → Train a model → Deploy

---

What you'll upload

Your dataset needs two types of data:

1. Assets (images)

Images are the visual data your model will learn from. Datature Vi supports a wide range of image formats including JPEG, PNG, TIFF, WebP, HEIF, and more.

Upload methods:

• Web interface — Drag-and-drop or file browser for quick uploads
• Vi SDK — Programmatic uploads for large datasets and automation

Learn more about uploading images →

2. Annotations (labels)

Annotations define what your model should learn to recognize—bounding boxes, phrases, or question-answer pairs depending on your task type.

Upload options:

• Import existing annotations — From COCO, YOLO, Pascal VOC, CSV, or Vi JSONL formats
• Create new annotations — Using the visual annotator or AI-assisted tools

Learn more about uploading annotations →

---

Upload workflow

Follow this sequence for best results:

📘

Important upload order

Always upload images before annotations. The annotation system matches labels to images by filename, so images must exist in the dataset first.

---

Choose your upload method

Web interface uploads

Best for: Small to medium datasets (under 1,000 images), visual workflows, one-time uploads

Features:

• Intuitive drag-and-drop interface
• Real-time progress tracking
• Visual file selection and preview
• No coding or technical setup required
• Background processing while you work

Get started:

• Upload images via web interface →
• Upload annotations via web interface →

SDK (programmatic) uploads

Best for: Large datasets (1,000+ images), automation, CI/CD integration, repeated workflows

Features:

• Efficient batch processing
• Advanced error handling and retry logic
• Integration with existing data pipelines
• Automated workflows and scheduling
• Precise control over upload behavior

Get started:

• Vi SDK Getting Started →
• Assets API Reference →
• Annotations API Reference →

---

Supported formats

Datature Vi supports a wide range of image formats (JPEG, PNG, TIFF, WebP, HEIF, and more) and annotation formats (COCO, YOLO, Pascal VOC, CSV, Vi JSONL).

Learn more:

• Supported image formats and specifications →
• Supported annotation formats with examples →

---

Best practices

Key tips:

• Use JPEG or PNG formats for images (under 10 MB recommended)
• Always upload images before annotations
• Ensure annotation filenames exactly match image filenames (case-sensitive)
• Upload in batches of 10-50 images for reliability
• For large datasets (1,000+ images), use the SDK instead of web interface

Detailed guidelines:

• Image upload best practices →
• Annotation upload best practices →

---

Troubleshooting

Having issues with uploads? Check the detailed troubleshooting guides:

• Image upload troubleshooting → — Slow uploads, unsupported formats, file size issues
• Annotation upload troubleshooting → — Missing annotations, format errors, filename mismatches

---

Next steps

Once your images and annotations are uploaded:

---

Detailed guides

• Upload assets (images) → — Complete guide to uploading images via web and SDK
• Upload annotations → — Import annotations from all supported formats

---

Related resources

• Create a dataset — Learn how to create datasets for different task types
• Manage datasets — Rename, delete, and organize your datasets
• Download data — Export datasets and annotations
• Vi SDK getting started — Get started with programmatic uploads
• Phrase grounding concepts — Learn about object detection tasks
• Visual question answering concepts — Learn about VQA tasks
• Upload assets — Step-by-step image upload guide
• Upload annotations — Import existing annotations
• Annotate data — Create annotations manually
• View dataset insights — Check upload progress and statistics
• Train a model — Use your data for VLM training
• Quickstart — End-to-end workflow tutorial