Label the unseeable,
with cinematic precision.
LabelX turns pathology slides, OPGs, CT and microscopy into pixel-perfect datasets. Draw polygons, boxes, lines and circles by pen, touch or mouse — or just click and let SAM 2 segment it. Every image is embedded into Qdrant and ranked by an active-learning loop, so you label what matters first.
- 5
- annotation tools
- SAM 2
- promptable segmentation
- Qdrant
- vector memory
- Active
- learning loop
One canvas, every tool
A complete toolbox for the microscopic.
Every primitive you need to describe biology — drawn by pen on a tablet, by finger on a touchscreen, or by mouse. Precise, keyboard-accelerated, and undo-friendly.
Click once. Get a perfect mask.
Drop a point or drag a box and SAM 2 proposes a full segmentation instantly. Add positive and negative prompts to refine, then commit the mask as an editable polygon. Ambiguous region? Ask for multiple mask hypotheses and pick the best.
+ point
positive
− point
exclude
box
region
mask
commit
Polygon
Trace intricate cell membranes vertex by vertex, with editable anchor points.
Rectangle
Fast bounding boxes for detection datasets and ROI selection.
Line & polyline
Measure structures, mark axes, trace ducts and vessels.
Circle & ellipse
Perfect for nuclei, follicles and round morphologies.
Pen, touch or mouse
Pressure-aware freehand and pointer-events for stylus and Wacom, tuned for tablets and touchscreens.
Layers & classes
Organize annotations into ontology classes with colors, visibility, lock and z-order.
Export anywhere
One click to COCO, LabelMe or mask PNGs — ready for training pipelines.
From pixels to datasets
A pipeline that closes the loop.
LabelX is not just a canvas — it's the full path from a raw image to a model-ready dataset, and back again.
Import
Drag in slides, OPGs, CT frames or microscopy — single files or whole folders.
Annotate
Draw polygons, boxes, lines and circles with pen, touch or mouse.
Segment
Prompt SAM 2 with a click or box and commit the mask as a polygon.
Embed
Each image is vectorized and cached in Qdrant — compute once, reuse forever.
Prioritize
Active learning ranks the most uncertain samples for the next round.
Export
Ship a clean COCO / LabelMe dataset straight into your training loop.
Segment Anything 2
Promptable masks, in a heartbeat.
Skip the tedium of tracing by hand. Guide the model with a click or a box, review the proposal, and refine — LabelX handles the heavy embedding compute and caches it so re-prompting is instant.
Point prompts
Positive & negative clicks
Box prompts
Drag a region to segment
Multi-mask
Pick from mask hypotheses
Video-ready
Track masks across frames
1 · Raw image
2 · Click prompt
3 · Instant mask
Label the 10% that teaches your model the most.
Connect a dataset to a model, run inference, and LabelX ranks every unlabeled image by how much it would improve the model. Spend your annotation hours where they actually move the metric.
Uncertainty sampling
Least-confidence, margin and entropy scoring surface the samples your model finds hardest.
Diversity & novelty
K-center selection over Qdrant embeddings avoids near-duplicates and finds new morphologies.
Bring your own model
Register any model endpoint, feed it your labels, and stream predictions back into the queue.
Labeling priority queue
hybrid strategygastric_2231.tiff
0.97
high entropy
opg_lower_087.png
0.94
low margin
ct_thorax_412.png
0.88
novel region
blood_smear_66.jpg
0.72
uncertain
histo_liver_09.tiff
0.41
confident
Est. labels saved this round
~ 63%
Built for biology
Every modality, one studio.
Start with biomedical images today — video annotation for motility and live-cell studies is on the roadmap.
Histopathology
Whole-slide & tiled H&E, IHC, cytology.
Dental OPG & X-ray
Panoramic radiographs, periapicals, bitewings.
CT & MRI slices
Windowed grayscale volumes, frame by frame.
Blood smears
Cell counting, morphology, classification.
Endoscopy & fundus
Lesion and structure delineation.
Motility video
Sperm & live-cell tracking with masks across frames.
Turn your images into intelligent datasets.
Open the Studio and annotate your first slide in seconds. No setup, no friction — just precision.