Annotate Objects

Guides

Object Detection

Annotate Objects

Draw bounding boxes around objects in your images. Accurate annotations are crucial for model performance.

Annotation Interface

Using the Web Annotation Tool

Open your dataset > Annotate tab
Select an image to annotate
Use the toolbar:
- Box Tool (B) - Draw bounding boxes
- Select Tool (V) - Edit existing boxes
- Zoom (Z) - Zoom in for precision

Drawing Bounding Boxes

Click the Box Tool or press B
Click and drag around the object
Select the appropriate label
Repeat for all objects in the image
Click Save or press Ctrl+S

Keyboard Shortcuts

Key	Action
`B`	Box tool
`V`	Select/Edit tool
`N`	Next image
`P`	Previous image
`D`	Delete selected box
`1-9`	Quick label select
`Ctrl+S`	Save
`Ctrl+Z`	Undo

Programmatic Annotation

from seeme import Client

client = Client()

# Get dataset items
items = client.get_dataset_items(
    dataset_id=dataset.id,
    version_id=version.id
)

# Add bounding box annotation
for item in items:
    # Example: detected objects from another system
    detections = your_detection_function(item.file_path)

    for det in detections:
        client.annotate(
            dataset_id=dataset.id,
            dataset_version_id=version.id,
            annotation={
                "item_id": item.id,
                "annotation_type": "bounding_box",
                "label_id": labels[det["class"]].id,
                "x": det["x"],
                "y": det["y"],
                "width": det["width"],
                "height": det["height"]
            }
        )

Import Existing Annotations

# From YOLO format
client.import_annotations(
    dataset_id=dataset.id,
    version_id=version.id,
    format="yolo",
    annotations_path="./labels/"
)

# From COCO format
client.import_annotations(
    dataset_id=dataset.id,
    version_id=version.id,
    format="coco",
    annotations_path="./annotations.json"
)

Annotation Guidelines

Box Placement Rules

Good annotation:
┌─────────────────┐
│   ┌─────────┐   │  Box tightly surrounds
│   │ [car]   │   │  the visible object
│   └─────────┘   │
└─────────────────┘

Bad annotation:
┌─────────────────┐
│ ┌─────────────┐ │  Box includes too much
│ │             │ │  background
│ │   [car]     │ │
│ └─────────────┘ │
└─────────────────┘

Handling Special Cases

Scenario	Approach
Partially visible	Annotate visible portion
Occluded objects	Annotate visible parts, or use “occluded” tag
Overlapping objects	Create separate boxes for each
Small objects	Zoom in for precision
Ambiguous objects	Use “difficult” or “uncertain” tags

Quality Control

Review Workflow

Random sampling - Review 10% of annotations
Inter-annotator agreement - Have multiple people annotate same images
Edge case review - Focus on difficult images
Label distribution - Check for class imbalance

Annotation Statistics

# Get annotation statistics
stats = client.get_dataset_stats(
    dataset_id=dataset.id,
    version_id=version.id
)

print(f"Total images: {stats['total_items']}")
print(f"Annotated: {stats['annotated_items']}")
print(f"Boxes per image: {stats['avg_annotations_per_item']}")

# Label distribution
for label, count in stats['label_counts'].items():
    print(f"  {label}: {count} boxes")

Data Augmentation Preview

SeeMe.ai applies augmentations during training. Preview how your annotations will be transformed:

# Preview augmentations
augmented = client.preview_augmentation(
    item_id=item.id,
    augmentation_type="flip_horizontal"
)

# Shows image with transformed bounding boxes

Best Practices

Consistency is key - Same rules for all annotators
Tight boxes - Minimize background, maximize object
Label everything - Don’t skip small or partially visible objects
Use labels consistently - “car” vs “Car” vs “CAR” are different!
Take breaks - Annotation fatigue leads to errors

Next Step

3. Train Model →

Prepare Your Object Detection Dataset Train Your Object Detection Model