Annotation Formats

SAM Annotator supports multiple annotation formats for exporting your segmentation data. This guide explains the different formats, their structure, and how to use them.

Supported Export Formats

SAM Annotator currently supports exporting annotations to the following formats:

1. COCO: The Common Objects in Context format
2. YOLO: You Only Look Once format
3. Pascal VOC: Pascal Visual Object Classes format

Exporting Annotations

You can export your annotations using the following keyboard shortcuts:

Format	Shortcut
COCO	E
YOLO	Y
Pascal VOC	V

All exports are saved to the exports/ directory within your category path, with a subdirectory for each format.

Format Details

COCO Format

COCO is a large-scale object detection, segmentation, and captioning dataset format. SAM Annotator exports annotations in the COCO format as a JSON file with the following structure:

{
  "info": { ... },
  "images": [
    {
      "id": 1,
      "file_name": "image1.jpg",
      "width": 640,
      "height": 480
    },
    ...
  ],
  "annotations": [
    {
      "id": 1,
      "image_id": 1,
      "category_id": 1,
      "segmentation": [[x1, y1, x2, y2, ...]],
      "area": 1000,
      "bbox": [x, y, width, height],
      "iscrowd": 0
    },
    ...
  ],
  "categories": [
    {
      "id": 1,
      "name": "person",
      "supercategory": "none"
    },
    ...
  ]
}

Use Cases:

• Compatible with COCO API and many deep learning frameworks
• Ideal for object detection and instance segmentation tasks
• Good for large datasets with multiple classes

YOLO Format

YOLO is a popular real-time object detection system. SAM Annotator exports segmentation annotations in the YOLO segmentation format:

# One file per image
# Each line: class_id x1 y1 x2 y2 ... xn yn
# Where coordinates are normalized (0-1)
0 0.507 0.425 0.531 0.421 0.529 0.437 0.506 0.441
1 0.322 0.661 0.359 0.654 0.356 0.674 0.320 0.681
...

This format is compatible with YOLOv5, YOLOv8 and other versions that support segmentation.

Use Cases:

• Real-time object detection and segmentation
• Deployment on edge devices
• Training custom YOLO models

Pascal VOC Format

Pascal VOC is a dataset for object class recognition. SAM Annotator exports annotations in Pascal VOC XML format:

<annotation>
  <folder>images</folder>
  <filename>image1.jpg</filename>
  <size>
    <width>640</width>
    <height>480</height>
    <depth>3</depth>
  </size>
  <object>
    <name>person</name>
    <bndbox>
      <xmin>100</xmin>
      <ymin>150</ymin>
      <xmax>200</xmax>
      <ymax>250</ymax>
    </bndbox>
    <polygon>
      <point>
        <x>120</x>
        <y>160</y>
      </point>
      <point>
        <x>180</x>
        <y>160</y>
      </point>
      ...
    </polygon>
  </object>
  ...
</annotation>

Use Cases:

• Compatible with older computer vision tools
• Standard format for many object detection datasets
• Easy to parse and inspect manually

Choosing the Right Format

The best format to export your annotations depends on your specific needs:

• COCO: Choose for machine learning frameworks that support COCO format, such as TensorFlow Object Detection API, Detectron2, or MMDetection
• YOLO: Best for YOLO-based models, particularly if you're using Ultralytics' YOLO implementations
• Pascal VOC: Good for compatibility with older systems or when XML-based formats are required

Format Conversion (Coming Soon)

Support for converting between different annotation formats directly within SAM Annotator is planned for a future release.

External Tools

For more advanced conversions or formats not yet supported by SAM Annotator, consider these external tools:

1. Roboflow - Supports multiple format conversions and dataset preprocessing
2. LabelMe - Another annotation tool with format conversion utilities
3. COCO-to-YOLO - Scripts for converting COCO JSON to YOLO format

Best Practices

1. Export Regularly: Export your annotations frequently to prevent data loss
2. Version Control: Keep track of different versions of your exported annotations
3. Metadata: Document any preprocessing or specific details about your exports
4. Validation: Verify that exported annotations are correctly loaded in your target application