AI-Powered Automation
YOLO + MobileSAM Integration
We integrate MobileSAM for assisted segmentation. The detection engine is fully agnostic: load your own YOLO weights in ONNX format to keep complete control over your models.
Supported YOLO Models
All models must be exported to .onnx format before loading. The engine reads tensor output directly — no framework dependencies required.
YOLOv5 and YOLOv7 use a different ONNX output format and are not compatible with the current inference engine. YOLOv10's end-to-end architecture removes the non-maximum suppression (NMS) post-processing step, reducing inference latency.
Native ONNX Inference
Local execution via ONNX Runtime. We avoid heavy frameworks like PyTorch or Ultralytics to maximize compatibility and efficiency on your hardware.
High-Speed Annotation
Fast Mode: An interface designed to reduce friction in repetitive tasks, enabling instant class assignments.
Versatile tools: Native support for Bounding Boxes, Polygons, Points, and Lines, always keeping precision under your control.
Quality Control & Data Ops
False Negative Review
A dedicated tool for catching model omissions. Automatically filters and reviews areas where the model lacks confidence, ensuring dataset integrity. As of v1.0.1, metric calculations run asynchronously in the background, so the interface stays fully responsive even across 5,000+ image batches.
Export-Time Augmentations
Apply geometric and color transformations dynamically on export. Prepare your data for training without altering your original files.
Real-Time Statistics
A dashboard with instant metrics on class distribution and labeling progress. Spot dataset imbalances before exporting.
More Under the Hood
Beyond annotation and AI tooling, LensLaber covers the full dataset lifecycle — from large-scale browsing to export-ready packaging.
Large Dataset Workflow
Thumbnail navigation, multi-selection, and fast image switching keep things responsive across 20,000+ image datasets.
Project Save/Load System
Save and restore full workflow state — annotations, review progress, filters, and history — using the native .lens project format.
Dataset Export
Customizable Train/Validation/Test splits, optional image inclusion, and single-file ZIP packaging across multiple annotation formats. Unlimited daily exports as of v1.0.1 — the previous 1,000-export cap has been removed.
Image Quality Control
Pre-export checks for resolution, aspect ratio, compression artifacts, and lighting issues, so problems get caught before training.
Configurable Shortcuts
Fully customizable keyboard controls let you tailor the annotation workflow to your own habits and speed.
Adaptive Cursor & Autosave
A context-aware cursor reflects the active tool in real time, while autosave continuously protects your progress in the background.
Ready to try it on your own dataset?
Download the beta for Windows or Linux and start annotating offline in minutes.
Download Beta →As a beta product, LensLaber is still evolving — we deeply appreciate any feedback you share along the way, as it directly shapes future releases.