Image and video annotation services
Bounding boxes, polygons, segmentation, keypoints, object tracking and captions, drawn to your label set and checked by a second annotator. Text inside images is transcribed by native readers of the script.
4 boxes. truck, truck, person, no-entry sign. Each box is tight to the visible edge of the object.
2 polygons. barrier and truck, traced point by point for instance segmentation.
- 食品 啟發 茶莊 KAI FAT
- 長興燕窩莊
- 啟發公司
- Toilets 公廁
A narrow Hong Kong street with shop signs in Traditional Chinese and two blue delivery trucks.
香港一條狹窄的街道，兩旁掛著繁體中文招牌，路上有兩輛藍色貨車。
- Label types
- Boxes, polygons, masks, keypoints, tracks and captions
- Data
- Photos, product images, documents, video and drone footage
- Output
- COCO, YOLO, Pascal VOC, CSV or labels in your platform
- Quality
- Second-annotator review and gold items on every batch
What image annotation is
Image annotation is marking what's in a picture, such as drawing a box around each car or tracing the outline of a road, so a computer vision model can learn to find those things on its own. Video annotation does the same across frames and keeps each object's ID as it moves.
Our particular strength is images with language in them: shop signs, labels, menus, forms and screenshots in Chinese, Arabic, Hindi and other scripts, transcribed and captioned by people who read them.
Annotation types
Drawn to your label set and guidelines, in your tool or ours.
Bounding boxes
Boxes drawn tight to each object, with the class and any attributes you need, such as occluded, truncated or parked.
Polygons and segmentation
Object outlines traced point by point for instance segmentation, or every pixel labeled by class for semantic segmentation.
Keypoints
Points placed on faces, hands, bodies or product parts in the order your skeleton defines, with visibility flags.
Video tracking
Objects followed frame by frame with a consistent ID, including when they leave the frame and come back.
Scene text and OCR
Text in photos, signs, packaging and screenshots boxed and transcribed exactly, by native readers of each script.
Captions in many languages
Image and video descriptions written directly in each language, so vision-language models learn how native speakers describe a scene.
How an annotation project runs
Edge cases get settled in the pilot, so production batches stay consistent.
Label set
You share the classes, attributes and examples. We list the edge cases we expect to meet.
Pilot batch
We annotate a sample, with part of it done twice to measure agreement between annotators.
Calibration
You review the pilot with us. We update the guidelines and lock them before scaling.
Your sign-off QAProduction
A second annotator reviews every batch, and gold items track each person's accuracy.
Delivery
Labels in your format or platform, with a batch report on agreement and open questions.
Formats and tools
- Data you can send
- JPG, PNG, TIFF, MP4 and MOV, or access to your storage bucket
- Label formats
- COCO JSON, YOLO, Pascal VOC XML, CSV, or masks as PNG
- Platforms
- Your labeling platform, or CVAT or Label Studio set up for your project
- Scripts we read
- Latin, Chinese, Japanese, Korean, Arabic, Hebrew, Devanagari, Thai and Cyrillic
Formats and platforms to confirm
How your images are handled
Photos and video often show people. These rules apply on every project. See how we handle data.
- Signed NDAs. Every annotator and reviewer works under one.
- Limited access. Only the people on your project can open your data.
- Faces and plates. Blurred or boxed for redaction when your project needs it.
- Deleted at the end. Unless you ask us to keep data for follow-up batches.
Related services
Questions about image and video annotation
What is image annotation?
Image annotation is marking objects and regions in images, with boxes, outlines, points or text labels, so a computer vision model can learn to recognize them. Video annotation does the same across frames and keeps each object's identity as it moves.
Which label types do you support?
Bounding boxes, polygons, semantic and instance segmentation masks, keypoints, object tracking in video, scene text transcription, and image and video captions.
Can you caption images in other languages?
Yes. Native speakers write captions directly in each language rather than translating English captions, so the descriptions read the way people in that market actually talk.
Can you transcribe text that appears in images?
Yes. We box each piece of text and transcribe it exactly, in the original script. That includes signs, packaging, menus, handwritten notes and screenshots in Chinese, Arabic, Devanagari and other scripts.
How do you check annotation quality?
Every batch gets a second-annotator review. Gold items with agreed answers are mixed in to score each annotator, and part of each batch is labeled twice to measure agreement. You get those numbers with every delivery.
Can you work in our labeling platform?
Yes. We work inside your platform under your access rules. If you don't have one, we set up a tool for the project and deliver labels in the format your training pipeline expects.
Start with a pilot batch
Send your label set and a sample of images or video. We annotate a pilot batch, report agreement and quote for the full set.
- Phone
- +44 20 3290 0116
- Phone hours
- Weekdays 9:30 to 16:00, weekends 9:30 to 12:00, UK time