How Much Do Image Annotation Services Cost?

Article by Rei Morikawa | July 16, 2019

Image annotation is the task of manually defining regions in an image and creating text-based descriptions of them. Image annotation is an important first step toward building training datasets for computer vision, which can then be used to build AI systems for content moderation, autonomous vehicles, face recognition, and more.

 

Where can you get cheap image annotation?

In this article, we’ll explore several options for where to get cheap image annotation services for AI and computer vision.

The first option is internal labeling, where you assign the image labeling tasks to an in-house data science team. The benefits to this approach are that you can track the progress and get predictable results. Internal labeling might be a good option if you only need a small sample of annotated images. If you have complex instructions, then it might actually be faster to annotate the images in-house, than to vet external annotators who might not understand all of the requirements.

If you only need a small sample of annotated images, you could also hire a small team of temp workers for in-house labeling. The disadvantages of this solution, however, are that it does not scale if you later realize that you need more images annotated, and that it comes with additional management costs of recruiting and training the team.

In many cases, the costs of internal labeling outweigh the benefits. Internal labeling consumes precious internal resources, lacks diversity, and takes a lot of time.

Most AI companies choose to outsource the time-intensive and manual task of image annotation. You can outsource image annotation tasks to general crowdsourcing and outsourcing companies, or specialized AI training data providers. If you reach out to the right outsourcing company, you can get the breadth of diversity and remove the management responsibilities of resource acquisition and quality assessment.

 

How is image annotation pricing calculated?

If you’re considering an outsourcing solution, the next thing you’ll probably want to know is about image annotation pricing. Here are some factors to consider when conducting your market research.

 

Image Annotation Quality Level

Some image annotation outsourcing companies have different pricing plans depending on the quality level that you’d like to receive. Image annotation quality includes ensuring that all objects in an image are correctly identified and labeled, and that the bounding boxes and key points are placed precisely on their targets. For example, Lionbridge’s image annotation pricing plans include the following.

  • Standard single pass: Each image is annotated by a single contributor.
  • Single pass + 10%: After single pass, Lionbridge’s project managers review a representative sample of the images to ensure high-quality image annotations. This image annotation option generally costs 10% more than the standard single pass option.
  • Double pass: Each image is annotated by two different contributors. The second contributor can see what the first contributor did, and make changes if they think it’s necessary. Double pass image annotation is generally priced at twice the cost of standard single pass.
  • Double pass blind: Each image is annotated by two different contributors who don’t see each other’s annotations. This way, even if one contributor makes a mistake, it won’t influence the second contributor. If the two contributors submit different image annotations, Lionbridge’s project managers will investigate the reason for that discrepancy, and make sure the error didn’t occur in other parts of the image annotation project.

 

Image Annotation Project Guidelines

For every image annotation project, it’s crucial to establish clear guidelines. You should build out a comprehensive document that gives clear instructions to the annotators, and clarifies exactly what kind of results you are looking to achieve for the image annotation project. The project guidelines should include what type of image annotation to use, and the best practices for annotation.

Lionbridge has worked on an image annotation project for robot dockers, so that robots can recharge their batteries. For that project, our customer had specific guidelines about how to annotate the dockers.

If you submit unusual or overly complicated guidelines to a crowdsourcing company, that will probably increase the cost of training their contributors.

 

Project Management Fee for Image Annotation

Project management includes organizing the budget and schedule for your image annotation task. If you’re using a crowdsourcing service, gathering crowdworkers with the right qualifications and training them would also fall under project management.

Project management fees also depend on what kind of tool you’d like to use for the image annotation task. At Lionbridge, if customers use our existing AI training data platform for their image annotation tasks, then the project management fee is quite low. We’ve built a platform where you can manage and optimize image annotation quality and capacity.

In other cases, our project management team will need to set up an integration and provide additional training so that our contributors can use a new platform. This would increase project management fees.

In many cases, the factors that could decrease image annotation pricing are fixed, especially the project guidelines. You might not be flexible about what kind of image annotation services you need for your machine learning project. But if you can use tools that are already integrated with Lionbridge’s AI training data platform, that can decrease costs significantly.

Lionbridge’s AI training data platform allows us to offer the lowest possible project management fees for image annotation. Our platform allows contributors to create, edit, delete, and submit image annotation tasks using bounding boxes. Project managers can also use our platform to create new image annotation projects, view progress, export the results in JSON format, and reject submitted image annotation tasks if needed.

IMAGE ANNOTATION
Lionbridge’s image annotation platform

 

Data Volume for Image Annotation

Image data volume is simply the number of images multiplied by the number of annotations per image that are requested. For example, you might have 1,000 images that need to be annotated with about 5 bounding boxes per image.

Some image annotation outsourcing companies, including Lionbridge, offer volume discounts for large projects.This is due to the economy of scale — using a specific team that is familiar with your image annotation project guidelines becomes cost efficient over time, because the team will become more and more familiar and adept with that project.

Sometimes it’s cheaper to order large image annotation tasks than small-scale tasks. We’ve seen that image annotation pricing is based on: (1) image data volume; (2) level of precision; and (3) project management fee. For most image annotation tasks, the project management cost is a fixed cost.

Let’s say that you order an image annotation task with the following cost breakdown.

  • Project management: $500
  • Cost per image: $1
  • Quality: Double pass

In this case, if you order 100 images to be annotated, you’ll get $200 of annotated images in relation to $500 project management fee. But if you order 10,000 images to be annotated, you’ll get $20,000 of annotated images in relation to the same $500 project management fee. If you think about it this way, large image annotation projects are relatively advantageous in terms of cost.

Looking for image annotation services?
The Author
Rei Morikawa

Rei writes content for Lionbridge’s website, blog articles, and social media. Born and raised in Tokyo, but also studied abroad in the US. A huge people person, and passionate about long-distance running, traveling, and discovering new music on Spotify.

    Welcome!

    Sign up to our newsletter for fresh developments from the world of training data. Lionbridge brings you interviews with industry experts, dataset collections and more.