Image Annotation: Bounding Boxes and Polygons, and When to Use Each
Computer vision models learn from images where people have marked what is in them. How you mark them, and how consistently, affects what the model learns.
We label images remotely from Chittorgarh, India, for AI teams in India and abroad. We start with a small pilot batch.
Our main page for this service: AI Data Annotation.
Boxes or polygons?
A bounding box is a rectangle around an object. It is fast and fine when you need to know what is where. A polygon follows the outline of the object. It takes longer, and it is the right choice when the exact shape matters, as with irregular objects or tight spaces.
- Boxes: counting, detecting and locating objects
- Polygons: precise shapes, areas and boundaries
- Many projects use both, on different classes
The cases that cause rework
Objects that are partly hidden, cut off by the edge of the picture, very small, or blurred need a rule. If the guide does not say what to do, different people choose differently and the labels disagree. We list these cases with example images in the guide.
How a pilot works
You send a small sample. We agree a written labelling guide with you, label the batch, and you check the result. We agree how accuracy will be measured and the target before we talk about volume.
Frequently asked questions
Which tool do you label in?
We can usually work inside the tool you already use. Tell us which one, and we say whether we have used it before.
Can you label video too?
Yes. See our video annotation guide.
How is our data kept private?
We agree who may see it and sign an NDA before the first file is shared.
How much does it cost?
It depends on the label type, the number of objects per image and the volume. We quote after seeing a sample.
Send us a sample
A few images and a description of the labels you need is enough to start. Call +91-86194 70448 or email contact@gtsltd.in. We work remotely, in English, with businesses in India and abroad.