Getting Started Right
Organize Before You Annotate
Before loading images into labelWise, organize them in folders by similarity (same scene, same object types, similar lighting). This helps you work in batches and reuse annotations.
- Group similar images together
- Ensure consistent naming conventions
- Remove any corrupted or duplicate images
- Verify all images are in supported formats (JPEG, PNG, WebP, etc.)
Efficient Annotation Techniques
Use Copy and Paste for Similar Objects
The most powerful workflow optimization in labelWise is the copy/paste feature:
Template workflow: Create a perfect annotation once, then copy and paste it across similar images. Adjust the position and size as needed.
- Annotate a car in the first image with precise boundaries
- Press
Ctrl/Cmd + C to copy
- Navigate to the next image with a similar car
- Press
Ctrl/Cmd + V to paste
- Drag to reposition and resize if needed
- Repeat for all similar images
Multi-Select for Batch Operations
Use Ctrl/Cmd + Click to select multiple annotations at once. This is perfect for:
- Copying multiple objects together
- Reviewing multiple annotations before moving on
- Deleting multiple incorrect annotations
- Annotate all objects on an image
- Hold
Ctrl/Cmd and click each annotation to select them all
- Copy all annotations with
Ctrl/Cmd + C
- Go to a similar image
- Paste all annotations at once
- Adjust individual positions as needed
Use Grid Guides for Alignment
Enable the grid guide (click the Guía button) to help align annotations consistently. This is especially useful for:
- Architectural images with straight lines
- Product photos requiring precise positioning
- Creating annotations of consistent size
GRID_SIZE constant) and helps maintain visual consistency across annotations.
Label Management
Bulk Label Creation
Use the bulk label textarea to create multiple labels at once. Enter labels separated by commas or line breaks:car, truck, bus, motorcycle
person, bicycle
traffic light, stop sign
Label Naming Conventions
Best practices for label names:
- Use lowercase (labelWise automatically converts to lowercase)
- Keep names short but descriptive (
car not automobile_vehicle)
- Use underscores for multi-word labels (
traffic_light not traffic light)
- Be consistent across your dataset
- Avoid special characters and emojis
Pre-Define Your Label Set
Create all your labels before you start annotating. This prevents:
- Typos and inconsistent naming
- Having to switch between drawing and label creation
- Accidentally creating duplicate labels with different names
Canvas Navigation
Zoom and Pan Effectively
For detailed work:
- Switch to Pan mode (click Mover button)
- Use the + button to zoom in up to 300%
- Drag the canvas to navigate
- Switch back to Draw mode to create annotations
- Use Reset to return to 100% zoom
Pro tip: Zoom in when working on small objects or fine details. The minimum box size is 10×10 pixels, so zooming helps with precision.
Two-View Workflow
Toggle between Canvas and CSV views to leverage different perspectives:
- Canvas view: Visual annotation, drawing, positioning
- CSV view: Precise numeric adjustments, bulk review, label changes
Use CSV view to:
- Quickly change labels without clicking through the canvas
- Enter exact pixel coordinates for perfect alignment
- Review all annotations across all images in one table
- Multi-select rows with
Ctrl/Cmd + Click
Export Habits
Save Early, Save Often
Critical: Export your CSV frequently. labelWise runs entirely in your browser with no server-side storage. If you close the tab or refresh the page, all annotations are lost.
- After completing each batch of 10-20 images
- Before taking a break
- Before closing the browser
- After any significant annotation session
Incremental Exports
Exported CSV files are named annotations-YYYY-MM-DD.csv. To avoid overwriting:
- Rename downloaded files immediately (add batch number, timestamp, etc.)
- Keep a backup folder with dated exports
- Merge CSVs in Excel or Python if working in multiple sessions
Annotation Quality Tips
Consistent Bounding Box Strategy
Decide on a consistent strategy for all annotations:
Tight boxes: Touch the object edges with minimal background
- Pros: More precise, better for segmentation models
- Cons: Time-consuming, harder on complex shapes
Loose boxes: Include some background margin
- Pros: Faster annotation, more forgiving
- Cons: Less precise, may include irrelevant context
Choose one strategy and stick to it across your entire dataset. Inconsistent bounding box styles can confuse ML models during training.
Handle Occlusions
For partially visible objects:
- Annotate the visible portion only (don’t guess the hidden part)
- Use consistent label names for partial views if relevant to your model
- Consider creating separate labels like
person_partial if occlusion is important
Edge Cases
Objects cut off by image boundary:
- Annotate the visible portion
- Extend box to image edge (coordinates are clamped automatically)
Overlapping objects:
- Annotate all objects, even if boxes overlap
- Z-order doesn’t matter in CSV export
- Use multi-select to manage overlapping annotations
Image Loading
For large datasets:
- Load images in batches of 50-100 at a time
- Remove completed images before loading the next batch
- Use consistent image sizes to maintain predictable performance
Browser Considerations
- Use Chrome or Edge for best performance
- Close unused tabs to free memory
- Disable unnecessary browser extensions
- For very large images (>4K resolution), consider resizing before annotation
Collaborative Workflows
Splitting Work Across Team Members
- By image set: Divide images into folders, each person annotates their folder
- By label type: Person A annotates vehicles, Person B annotates pedestrians
- By review phase: Person A annotates, Person B reviews and corrects
Merging Annotations
Since labelWise uses CSV format, you can merge work:
import pandas as pd
# Load multiple CSV exports
df1 = pd.read_csv('annotations-person-a.csv')
df2 = pd.read_csv('annotations-person-b.csv')
# Merge and deduplicate
merged = pd.concat([df1, df2]).drop_duplicates()
merged.to_csv('annotations-merged.csv', index=False)
Keyboard-First Workflow
For maximum speed:
- Pre-create all labels
- Select first label
- Load images
- Draw annotations on first image
Ctrl/Cmd + C to copy all annotations- Arrow key or click to next image
Ctrl/Cmd + V to paste- Adjust positions
- Repeat from step 6
- Export CSV every 20 images
This workflow can achieve 10-15 images per minute for repetitive annotation tasks with consistent object placement.