How do you calculate annotation hours?

Divide the total images by annotation throughput to get base hours, then add the rework and QA allowance. With 5,000 images at 40 images/hr, base time is 125 hours; a 20% allowance adds 25 hours for a total of 150 annotation hours.

What is a realistic annotation throughput?

It depends entirely on the task. Simple bounding boxes can run well above 40 images/hr, while pixel-accurate segmentation of subtle defects can drop below 10. The example's 40 images/hr fits moderate-complexity defect boxing; set yours from a timed pilot.

Why include a rework and QA allowance?

First-pass labels are never final — they get reviewed, corrected, and sometimes redone. A 20% allowance turns the 125 base hours into 150, capturing the review loop that otherwise blows up the schedule late.

How do I convert annotation hours into a schedule?

Divide total hours by annotator count and working hours per day. The 150-hour example is roughly four days for one annotator, or one to two days with a small team — before accounting for ramp-up and breaks.

What is a good rework allowance percentage?

For straightforward tasks 10-20% is typical; for ambiguous defect classes or new annotators it can reach 30-50%. The 20% in the example is a reasonable middle for a trained team on a defined defect spec.

In-house vs vendor annotation — how does this help?

The hour estimate lets you compare apples to apples: multiply total hours by your loaded internal rate versus a vendor's per-hour or per-image quote. At 150 hours, even a modest rate difference makes the build-versus-buy decision clear.

Machine Vision & Industrial Inspection AI calculator

Annotation Workload Calculator

Annotation workload estimates the labor hours needed to label a machine vision training dataset, including the rework and QA review that real labeling projects always incur. ML leads and quality teams use it to budget headcount and schedule before kicking off a data-labeling effort, whether in-house or with a vendor. Raw image counts divided by a labeling speed undercount the true effort because first-pass labels get reviewed and corrected. This calculator builds that rework allowance in so the estimate survives contact with reality.

What this calculator does

Estimate total annotation labor hours needed to label a set of inspection images for AI model training, based on image count, annotation throughput, and a rework and QA allowance.
Use it when planning an AI inspection project and you need to size the annotation effort before assigning annotators or budgeting for an annotation service.
It computes total annotation hours by dividing images by labeling throughput and adding a rework and QA review allowance.

Formula used

Base annotation time = total images / annotation throughput
Total annotation hours = base time x (1 + rework and QA allowance / 100)

Inputs explained

Total images to annotate:
Annotation throughput:
Rework and QA review allowance:

How to use the result

Use it when planning a labeling campaign, quoting a vendor, or sizing internal annotator headcount against a delivery deadline.
It uses a single average throughput; complex segmentation or multi-class defect labeling is far slower than simple bounding boxes, so set throughput to match the actual task.

Common questions

How do you calculate annotation hours? Divide the total images by annotation throughput to get base hours, then add the rework and QA allowance. With 5,000 images at 40 images/hr, base time is 125 hours; a 20% allowance adds 25 hours for a total of 150 annotation hours.
What is a realistic annotation throughput? It depends entirely on the task. Simple bounding boxes can run well above 40 images/hr, while pixel-accurate segmentation of subtle defects can drop below 10. The example's 40 images/hr fits moderate-complexity defect boxing; set yours from a timed pilot.
Why include a rework and QA allowance? First-pass labels are never final — they get reviewed, corrected, and sometimes redone. A 20% allowance turns the 125 base hours into 150, capturing the review loop that otherwise blows up the schedule late.
How do I convert annotation hours into a schedule? Divide total hours by annotator count and working hours per day. The 150-hour example is roughly four days for one annotator, or one to two days with a small team — before accounting for ramp-up and breaks.
What is a good rework allowance percentage? For straightforward tasks 10-20% is typical; for ambiguous defect classes or new annotators it can reach 30-50%. The 20% in the example is a reasonable middle for a trained team on a defined defect spec.

Last reviewed 2026-05-12.