How This Helps

Export datasets for use in training pipelines, labeling platforms, analytics workflows, or external review. You can export everything or just what you need.

​
Export a Dataset

1

Open the Dataset

Go to the Dataset Exploration page.
2

Click 'Export'

Click the Export button in the top-right corner.
3

Name the Export

Provide a descriptive name for your export file.
4

Choose Export Scope

Select the scope—entire dataset, filtered view, or specific selections.
5

Pick a Format

Choose from supported formats:
  • JSON– available now
  • COCO– coming soon
6

Include Images (Optional)

Toggle Include image files to include up to 5,000 images in your export.
7

Export

Click Export File to download.

​
Export Options

​
Entire Dataset

Exports all items in the dataset, regardless of filters or selections.

​
Matching Filters

Exports only the items that match your currently applied filters, such as issue types, tags, or labels.

​
Current Selection

Exports only manually selected items—images, objects, or clusters.

​
JSON Export Format

Exported data is structured into a .json file with two sections:

​
1. info

General metadata about the dataset and export: 
"info": {
  "dataset": "Name",
  "dataset_creation_time": "2025-04-02T12:15:02.510168",
  "dataset_url": "http://app.visual-layer.com/dataset/.../data",
  "description": "Exported from Visual Layer",
  "export_time": "2025-04-02T14:31:30.931324",
  "exported_by": "User_Name",
  "schema_version": "1.1",
  "total_media_items": 10000000
}

​
2. media_items

An array of media items (images or video frames), each with all available metadata: 
{
  "media_id": "<uuid>",
  "media_type": "image",  // or "video_frame"
  "file_name": "example.jpg",
  "file_path": "example.jpg",
  "file_size": "48.96KB",
  "height": 360,
  "width": 360,
  "url": "http://...",
  "cluster_id": "<uuid>",
  "uniqueness_score": 0.85,
  "metadata_items": [ ... ]
}

​
Metadata Items

Each media item includes a list of metadata_items. Below are the possible types:

​
1. type: "issue"

Used to represent a detected problem: 
{
  "type": "issue",
  "properties": {
    "issue_type": "duplicates",
    "confidence": 0.85,
    "duplicate_group_id": "<uuid>",
    "duplicate_threshold": 0.88
  }
}

​
2. type: "user_tag"

Tags applied by a user to a media item: 
{
  "type": "user_tag",
  "properties": {
    "tag_name": "spider",
    "assigned_date": "2025-04-02T12:34:00.189534"
  }
}

​
3. type: "video_info"

Provides context on a video frame: 
{
  "type": "video_info",
  "properties": {
    "frame_number": 17,
    "video_id": "<uuid>",
    "timestamp": "00:00:00.567"
  }
}

​
4. type: "object_label"

Describes an annotated object (bounding box), including nested issues: 
{
  "type": "object_label",
  "properties": {
    "bbox": [x, y, width, height],
    "category_name": "Joey",
    "source": "User",
    "original_id": -1,
    "uniqueness_score": 0.85,
    "url": "http://...",
    "metadata_items": [
      {
        "type": "issue",
        "properties": {
          "issue_type": "blur",
          "confidence": 0.81
        }
      }
    ]
  }
}

​
5. type: "image_label"

A classification label applied to the entire image: 
{
  "type": "image_label",
  "properties": {
    "category_name": "car",
    "source": "Model",
    "original_id": 1234
  }
}

​
6. type: "caption"

{
  "type": "caption",
  "properties": {
    "text": "A dog playing in the field",
    "source": "AI model"
  }
}
  • Duplicates often include duplicate_group_id and threshold values.
  • Object-level issues are nested inside object_label blocks.