AnnoVis is a lightweight image-annotation extension for VS Code / Cursor. It supports multiple annotation types including object detection (bounding boxes) and image classification directly inside the editor, saving annotations as JSON alongside your project.
- Draw, move and resize bounding boxes on images.
- Assign custom labels with unique colors.
- Maintain multiple annotation projects per workspace.
- Classify entire images with multiple labels.
- Assign confidence scores to each classification.
- Support for multi-label classification.
- Real-time classification results display.
- Draw precise polygon outlines around object instances.
- Click-based polygon drawing with visual feedback.
- Support for complex shapes and detailed boundaries.
- Edit and refine polygon vertices.
- Mark specific points and joints on objects.
- Create skeleton connections between related keypoints.
- Support for human pose estimation and object landmarks.
- Visual feedback for keypoint placement and connections.
- Support for common image formats (PNG, JPG, GIF, BMP, TIFF).
- Saves annotations as plain JSON for easy post-processing.
- Multiple annotation projects per workspace.
- Shared label management across annotation types.
- Project-based workflow with type selection.
- Use AnnoVis: Set Project command from the Command Palette or right-click menu.
- Select + New Project to create a new project.
- Choose the project type:
- Object Detection: For drawing bounding boxes around objects
- Image Classification: For classifying entire images with labels
- Instance Segmentation: For drawing precise polygon outlines around object instances
- Keypoint Detection: For marking specific points and joints on objects
- Enter a project name.
- The project is now ready for annotation.
- Quick Annotation: Right-click an image file and choose Annotate Image with AnnoVis.
- This uses your current active project if you have one
- Only shows project selection if you don't have an active project
- Select Different Project: Right-click an image file and choose Annotate Image with AnnoVis (Select Project).
- Always shows project selection to let you switch projects
- The annotation interface will open based on your project type:
- Object Detection: Canvas with drawing tools for bounding boxes
- Image Classification: Label selection interface with confidence scoring
- Instance Segmentation: Canvas with polygon drawing tools for precise outlines
- Keypoint Detection: Canvas with keypoint placement tools and skeleton connections
- Projects are automatically saved and remembered.
- Each project has its own set of labels and settings.
- Switch between projects using:
- AnnoVis: Set Project command from Command Palette
- Annotate Image with AnnoVis (Select Project) when annotating
- Legacy Project Handling: If you have old projects without type information, the extension will ask you to specify whether they are Object Detection or Image Classification projects.
- Visualizing existing files: Right-click any annotation/classification JSON file and select "Visualize Annotation with AnnoVis" -> the correct project and interface will open automatically based on the file's metadata.
- Object Detection: Annotations stored in
/.annovis/annotations/[project-name]/ - Image Classification: Classifications stored in
/.annovis/classifications/[project-name]/ - Instance Segmentation: Instance annotations stored in
/.annovis/instances/[project-name]/ - Keypoint Detection: Keypoint annotations stored in
/.annovis/keypoints/[project-name]/ - Project Settings: Project configuration in
/.annovis/projects/[project-name]/project.json
Annotations are saved as JSON files with metadata and bounding box coordinates (normalized 0-1):
{
"metadata": {
"projectName": "My Detection Project",
"projectType": "object-detection",
"imageName": "image.jpg",
"created": "2024-01-01T12:00:00.000Z",
"version": "1.0"
},
"annotations": [
{
"label": "person",
"x": 0.1,
"y": 0.2,
"width": 0.3,
"height": 0.4
}
]
}Classifications are saved as JSON files with metadata and labels with confidence scores:
{
"metadata": {
"projectName": "My Classification Project",
"projectType": "image-classification",
"imageName": "image.jpg",
"created": "2024-01-01T12:00:00.000Z",
"version": "1.0"
},
"classification": {
"labels": [
{
"name": "outdoor",
"confidence": 0.9
},
{
"name": "nature",
"confidence": 0.7
}
],
"timestamp": "2024-01-01T12:00:00.000Z"
}
}Instance annotations are saved as JSON files with metadata and polygon coordinates (normalized 0-1):
{
"metadata": {
"projectName": "My Instance Project",
"projectType": "instance-segmentation",
"imageName": "image.jpg",
"created": "2024-01-01T12:00:00.000Z",
"version": "1.0"
},
"annotations": [
{
"type": "polygon",
"label": "person",
"points": [
{"x": 0.1, "y": 0.2},
{"x": 0.3, "y": 0.15},
{"x": 0.35, "y": 0.6},
{"x": 0.05, "y": 0.65}
]
}
]
}Keypoint annotations are saved as JSON files with metadata, keypoint coordinates (normalized 0-1), and skeleton connections:
{
"metadata": {
"projectName": "My Keypoint Project",
"projectType": "keypoint-detection",
"imageName": "image.jpg",
"created": "2024-01-01T12:00:00.000Z",
"version": "1.0"
},
"annotations": [
{
"type": "keypoint",
"id": "1704110400000.123",
"label": "head",
"x": 0.5,
"y": 0.3,
"visibility": true
},
{
"type": "keypoint",
"id": "1704110400001.456",
"label": "left_shoulder",
"x": 0.4,
"y": 0.4,
"visibility": true
}
],
"connections": [
{
"from": "1704110400000.123",
"to": "1704110400001.456"
}
]
}Project settings are stored in project.json:
{
"name": "My Detection Project",
"type": "object-detection",
"labels": [
{
"name": "person",
"color": "#e6194b"
}
],
"created": "2024-01-01T12:00:00.000Z"
}Note: The type field can be "object-detection", "image-classification", "instance-segmentation", or "keypoint-detection".
No external dependencies — works anywhere VS Code or Cursor runs (Windows, macOS, Linux).
• Initial preview release with Object Detection support. • Added annotation type selection for Image Classification. • Implemented full Image Classification functionality with confidence scoring. • Improved workflow: project type selection during creation, direct annotation based on project type. • Fixed critical bug: Project type corruption that was causing Image Classification projects to become Object Detection projects. • Enhanced project management: Project files now always contain type information, with automatic handling of legacy projects. • Improved workflow: Smart project selection that uses current project when available, with explicit option to switch projects. • New Feature: Instance Segmentation support with polygon-based annotation for precise object outlines. • New Feature: Keypoint Detection support with point-based annotation and skeleton connections for pose estimation.