Image annotation services for every pixel your model needs Boxes, masks, and points drawn by agents and checked by trained people
Image labeling services for computer vision models Damco draws the shapes your model learns to see, to one guide
How our image annotation outsourcing works From your sample images to annotated batches in your format
- Week 1Write the guideWe annotate a sample with you and record every rule and edge case.
- Week 2Run a pilotYou check a pilot batch, and we fix what the guide left unclear.
- From week 3Annotate at volumeAgents draw the first shapes, and annotators correct and finish them.
- Every batchReview and sign offSenior annotators audit a sample, and the batch owner signs off.
When outsourcing image annotation makes sense And when a model-assisted tool on its own is enough
A good fit when
- You need thousands of images annotated each week to one standard
- Your images are cluttered, partly hidden, or shot in poor light
- You want annotations in COCO, Pascal VOC, or YOLO, ready to train
Look elsewhere when
- Your images are simple, and auto-labeling already gets them right
- Only your own staff may view the images, by contract or law
- You need a few hundred images annotated once
Image annotation techniques in detail What each technique teaches a model, and when we use it
2D bounding boxes and 3D cuboids
A bounding box locates an object with a rectangle, and a cuboid adds depth so the model can judge size and direction. We use them for products on a shelf, vehicles on a road, and parcels on a belt.
Boxes are fast and cheap to draw. They're the right choice when the model must find an object, not trace its outline.
Semantic segmentation
Semantic segmentation assigns every pixel to a class, such as road, sidewalk, car, or sky. Instance segmentation goes one step further and tells two cars apart.
It takes longer than boxes, so we use it when the model needs exact edges, as in driving scenes, crop fields, or dents on a car panel.
Lines and splines
Lines and splines trace thin shapes that a box can't hold, such as lane markings, curbs, power lines, and cracks. Vehicle perception models learn to stay in lane from thousands of them.
Polygon annotation
Polygons outline irregular objects point by point, such as trees, buildings, street signs, logos, and watermarks. They fit tighter than a box, with far less effort than a full mask.
Keypoint and skeletal annotation
Keypoints mark exact positions on an object, such as the corners of an eye or the tips of the fingers. Skeletal annotation joins keypoints into a body, so a model can follow posture and movement in fitness, retail, and safety footage.
Image annotation for every industry
We annotate images for ecommerce catalogs, insurance claims, security and surveillance, agriculture, manufacturing inspection, and automotive. The guide for each says what counts as the object, what counts as damage, and what to do when it's hidden.
Questions about image annotation services What teams ask before they outsource image annotation
How do you ensure the quality and accuracy of image annotations?
Senior annotators review a sample of every batch, and automated checks catch shapes that break the guide, such as boxes that cut off part of an object. We track agreement between annotators to see where the guide needs work. The rules come from a guide you approve.
What file formats do you support for image annotation input and output?
We take JPEG, PNG, TIFF, and BMP images, among others. We deliver annotations as COCO JSON, Pascal VOC XML, YOLO TXT, or CSV, or in the format your pipeline already reads.
How do you handle ambiguous or edge cases in images?
Annotators flag an image they can't decide instead of guessing. A senior annotator decides it and writes the reason into the guide, and when the guide can't settle it, we ask you. Your answer becomes an example for everyone on the batch.
What is your pricing structure for image annotation outsourcing?
We price per image, per object, or per hour, depending on the technique and the precision you need. Boxes cost less than masks. The price comes from your pilot batch, so it reflects your real images.
Send us 50 images and the objects to find
We'll reply with the technique we'd use, how we'd run the pilot, and the price per image.