Data annotation services: Lithuania, Baltics, EU
Data annotation and labelling services
AKYS trains its own custom AI models on its own multisensor capture, which means it annotates data continuously; that is what training a model actually consists of. That capability is offered as a service: send us your data and we label it. The difference from a generic labelling vendor is what the data can be. LiDAR point clouds, 11K 360° panoramas, thermal and stereo depth are the formats AKYS works in every day, not exceptions it has to accommodate.
Annotation as a by-product of building models, not a side business
Most data labelling is sold by companies whose only product is labels. AKYS arrives at it from the other direction. It builds custom AI models for European road assets and for a specific city's own assets, and those models had to be trained on annotated multisensor data: road signs, potholes, cracks and asphalt damage, faded, worn or missing lane markings, poles, crosswalks, street lights, vegetation risk, each tagged with GPS coordinates. Alongside that: lane detection, semantic segmentation separating roadway from pavements, buildings and vegetation, spatial motion tracking of vehicles, pedestrians and cyclists, and AI 3D depth reconstruction.
None of that is possible without annotating the underlying data first. So the annotation is not a service bolted onto the business; it is the thing the business already does, offered to organisations that have their own data and need the same work done on it.
Beyond flat photos
Point clouds, spherical panoramas, thermal and stereo depth: the data types AKYS captures and labels for its own models.
Coordinates, not just pixels
AKYS's own detections carry GPS coordinates, so georeferenced data is a familiar input rather than a complication.
COCO-formatted
The same format AKYS uses for its own multisensor AI training datasets on marketplace.akys.ai.
Anonymise first
Faces and licence plates can be blurred by AI before annotation, in line with EU GDPR.
Data types we annotate
These are the formats AKYS works with natively, because they are the formats its own rig produces and its own models consume. Scope, class list and deliverables are agreed per project.
| LiDAR point clouds | .las / .laz: point and surface classes, objects delimited in three dimensions |
|---|---|
| 360° panoramas | 11K spherical imagery, the flagship output of the AKYS rig (Insta360 Titan, 11K, 10-bit, 8 × Micro Four Thirds) |
| Thermal | LWIR thermal imagery |
| Stereo depth | ZED 2i stereo depth: depth perception and spatial tracking data |
| Flat imagery | Standard images and video frames, regions and objects marked per your class list |
| Geospatial layers | GeoJSON / CSV: annotations tied to coordinates where the source data carries them |
| Output format | COCO-formatted AI training data |
| Class list | Yours, agreed before the work starts |
| Privacy | Optional AI blurring of faces and licence plates first: blurring service |
| Ready-made datasets | COCO-formatted multisensor AI training datasets: marketplace.akys.ai |
| Pricing | Quoted per project: [email protected] |
Who this is for
Annotation is only worth buying when the data is already hard to label in-house. These are the situations where that tends to be true.
AI, robotics and autonomy teams
Perception stacks that need labelled point clouds, depth and imagery together rather than a single flat modality. COCO output drops into a training pipeline.
Mapping and survey companies
You already produce .las and .laz at volume. Turning that geometry into training data is a different discipline from producing it, and it is the one AKYS does daily.
Research groups
Multisensor and geospatial datasets that need a consistent, documented class list before anyone can publish results off them.
Municipalities building their own models
AKYS already trains models for a specific city's assets. If you want to own the model, the annotated dataset underneath it is the part that has to exist first.
Teams with a privacy problem
Street-level or public-space data that cannot be sent anywhere until faces and licence plates are gone. Blur first, annotate after (see blurring service).
Teams who would rather buy than label
If the data does not have to be yours specifically, ready-made COCO multisensor datasets are on marketplace.akys.ai.
What annotating multisensor data actually involves
Labelling a photograph is a two-dimensional problem: an object is somewhere in a rectangle. Labelling a point cloud is not. The object exists in space, it is sampled unevenly, it is occluded from some angles and not others, and a class decision made on one side of it has to hold on the other. A 360° panorama has its own version of the problem: there is no frame edge, geometry distorts toward the poles, and the same object can appear at the seam.
This is why the modality matters more than the volume. AKYS's own models (lane detection, semantic segmentation, spatial motion tracking, AI 3D depth reconstruction) were trained on exactly this kind of data, which is the reason the annotation work is offered here at all. What gets labelled, and to what class list, is decided per project rather than pushed through a fixed template.
Frequently asked questions
What is data annotation, and what does AKYS annotate?
Data annotation is the work of adding machine-readable labels to raw data so a model can learn from it: marking where an object is in an image, which points in a LiDAR cloud belong to a surface, which pixels are roadway rather than pavement. AKYS annotates data belonging to other organisations, and the data does not have to come from an AKYS drive. The types handled natively are the ones AKYS captures and labels for its own models: .las and .laz point clouds, 11K 360° panoramas, LWIR thermal, ZED 2i stereo depth, and GeoJSON or CSV inventories.
Why use AKYS instead of a generic data labelling vendor?
Because AKYS is not primarily a labelling vendor; it is a company that trains its own custom AI models on its own multisensor capture, and annotation is what that requires. Its models detect road signs, potholes, cracks and asphalt damage, faded or missing lane markings, poles, crosswalks, street lights and vegetation risk, each with GPS coordinates, alongside lane detection, semantic segmentation, spatial motion tracking and AI 3D depth reconstruction. That means the data types are familiar rather than foreign: point clouds, spherical panoramas, thermal and stereo depth, with geospatial context attached. A vendor set up for flat photographs is being asked to learn all of that on your project.
Can AKYS annotate LiDAR point clouds and 360° panoramas?
Yes, these are the formats AKYS works with as a matter of course. Point clouds arrive and leave as .las or .laz, and 11K 360° panoramas are the flagship output of the AKYS rig, so a spherical frame is not an unusual input. Annotation of these data types covers what the geometry allows: regions and objects marked in imagery, classes assigned to points and surfaces in a cloud, objects delimited in three dimensions rather than in a flat frame. What is actually labelled, and to which class list, is agreed per project.
What format is the annotated data delivered in?
COCO-formatted AI training data. It is the format AKYS uses for its own multisensor datasets (LiDAR point clouds, 11K 360° vision and thermal), which are published at marketplace.akys.ai, so it is a working pipeline rather than an export written for one client. Geospatial layers can accompany the annotations as GeoJSON or CSV where the source data carries coordinates. If your training stack needs a different class list or structure, say so before the work starts.
Can data be anonymised before annotation, and what does it cost?
Yes. AKYS's own AI pipeline blurs faces and vehicle licence plates before storage in line with EU GDPR, and the same blurring is offered as a standalone service, so street-level or public-space data can be anonymised first and annotated after. Pricing is quoted per project, because a set of flat images and a georeferenced multisensor point cloud are not the same job. Send a sample of the data, the classes you need and any constraints to [email protected] and you will get a quote for that scope.
Get a quote for your annotation project
Tell us what the data is, what classes you need out of it, and what your training stack expects. Pricing is quoted per project.