Polygon Segmentation Annotator — CVAT & Segment Anything

About OpenTrain

OpenTrain is the #1 platform for finding and building careers in AI training and data labeling. We connect people to hands-on projects that teach models how to see, hear, and understand the world.

OpenTrain AI is the hiring and contracting organization for every role. We hire worldwide contributors as contractors for short-term and ongoing projects—no office required.

About AI Training Work

AI training (also called data labeling or annotation) is the human side of building machine intelligence: people prepare and verify examples that teach models to recognize objects, segment images, transcribe audio, and more. This work is remote, flexible, and a fast-growing way to work in tech.

Projects vary in skill needed; many are entry-level and only require attention to detail and basic familiarity with the tooling. Specialist tasks pay more for domain knowledge or advanced skills.

The Role

You will create polygon segmentations on recycling images, using CVAT alongside the Segment Anything interactor tool. Polygons are required for objects that exceed a specified size threshold and should separate individual items when multiple objects appear in one frame.

This is an entry-level contractor role, open worldwide. You will be paid hourly at the rate stated below and begin with a small paid trial to demonstrate speed and accuracy.

• Data type: images (recycling / waste items).
• Label types: object detection, polygon segmentation.
• Labeling software: CVAT (use Segment Anything interactor where applicable).
• Employment type: contractor, remote, worldwide.

What You'll Do

Perform careful polygon annotations to outline individual objects in a scene. Use CVAT and the Segment Anything interactor to speed up segmentation while ensuring polygons accurately separate distinct items.

Follow size-threshold rules: only annotate objects above the provided size threshold, and consistently distinguish unique objects when multiple items overlap or are adjacent.

• Open images in CVAT and run Segment Anything interactor where allowed to propose masks.
• Refine or redraw polygon outlines so each qualifying object is accurately represented.
• Label object boundaries cleanly to support downstream model training.
• Adhere to annotation guidelines for size thresholds, object separation, and consistency.

Requirements

This project is entry-level but requires attention to detail and the ability to distinguish individual objects in cluttered frames. Familiarity with CVAT and the Segment Anything tool is advantageous but not strictly required.

There are no additional formal qualifications. You must be able to work as an independent contractor and complete a short paid trial task.

• Attention to detail and visual discrimination skills.
• Familiarity with CVAT and Segment Anything interactor is a plus.
• Reliable internet connection and a computer capable of running CVAT.
• Willingness to complete a small paid test assignment to verify speed and accuracy.

Pay, Schedule, and Engagement

Pay type: hourly contractor work at USD 5.00 per hour. Work is remote and flexible; specific hours are not required by default.

You will start with a small amount of paid work as a trial. Successful contributors who meet speed and accuracy requirements may receive additional work opportunities.

• Hourly rate: $5.00 USD per hour.
• Work arrangement: contractor, remote, worldwide.
• Initial assignment: small paid test to evaluate speed and quality.

How It Works

Apply and you will receive a short paid trial assignment with clear guidelines and target accuracy/speed expectations. Complete the trial in CVAT using the Segment Anything interactor when instructed.

If you meet the project's quality and pace standards, you may be offered additional tasks. Feedback is provided so you can improve and qualify for more work over time.

• Start with a small paid trial assignment.
• Follow provided annotation guidelines exactly and ask clarification questions when needed.
• Consistently hit quality and speed targets to be considered for more work.