Image Annotation Jobs Summary
If they hire you for image annotation jobs, your task is to annotate or add labels to an image. Your annotations will provide computers with vision model information about what a particular image shows. Image annotation is an important part of artificial intelligence (AI).
The amount of labels on every image differs depending on a project. Some will require only one label that can represent what an entire image contains. There are also projects that call for multiple labels or tags since an image contains many objects.
Education and Training
Creating annotated images requires three things: the images, the person to annotate the images (you) and a platform on which you do the annotation.
The first step involves recruiting and training people to complete annotation projects and tasks. They require a master’s degree in some AI careers but not in image annotation. To qualify as an image or data annotator, there are really no specific educational requirements, although employers might prefer college graduate applicants. Still, you can find job ads with lower academic degrees, such as in Figure 1.
In this job post, the applicant needs at least a senior high school diploma. In most cases, however, annotators need thorough training on the guidelines and specifications of every project. Every company has its specific requirements for image annotation.
Once you have completed your training, you will then have access to a platform where you may have many images to annotate. The platform is a program with all necessary tools you need for the specific annotation assigned to you.
This video shows the use of a platform for labeling and annotating images.
Types of Image Annotation
As you have read earlier, a platform will provide you the tools you will use to perform a particular annotation. Therefore, know that there are different annotations you might need to perform.
1. 2D and 3D Bounding Boxes
With the 2D bounding boxes, you need to draw a box all over the image that you need to annotate within the image. Sometimes the target objects in an image are the same. The instructions could be to draw a box around every bicycle in an image.
There are also times when an image has different target objects with the instructions to draw a box around every car, bike, and pedestrian. Afterward, you will choose from a list of labels that are suitable attributions to the object inside the box.
They also call 3D bounding boxes as cuboids. Although similar to 2D bounding boxes, cuboids can also show the (approximate) depth of the target objects. You need to draw the boxes encircling the target object and to put anchor points at the edges of the object. There are times you cannot see a portion of the target object, such as when another object blocks it, so you will estimate the location of this blocked portion where you need to place the anchor points.
2. Image Classification
This process involves giving just one label to an entire image, such as labeling the animal or flower. They will give you images of animals or flowers, and your task is to categorize or classify them according to their types or species. A computer vision model will use the annotated image data to determine the unique characteristics of that animal.
3. Lines and Splines
This annotation involves labeling straight/curved lines on an image. Your task could include annotating sidewalks, power lines or lanes, and other boundary indicators. They also use the annotated images in drones’ trajectory planning. They use these lines and splines in a variety of cases involved in autonomous vehicles, drones, warehouse robotics and more.
There are target objects with irregular shapes. Bounding boxes are not suitable for these images, so they require polygons instead. You will plot points on every vertex of the object. Thus, no matter what shape the target object has, you can still annotate its exact edges.
5. Semantic Segmentation
Polygons, cuboids, and 2D bounding boxes all deal with annotation of individual objects inside the image. Semantic segmentation involves the annotation of each pixel within that image. Instead of a list of objects to annotate, you will have a list of segment labels into which you need to divide the image.
Suppose you are working on the traffic images for use by autonomous vehicles, your semantic segmentation task is to segment an image by cars, pedestrians, sidewalks, and buildings. A unique color code will indicate each segment. You will draw a line around the pixels and then choose the appropriate label.
Skills and Attributes
As a data annotator, you will help in categorizing contents. You could work on various images to assess their contents and attach tags appropriate to the content. You need to review the material you are working on to make sure nothing is amiss. Scrutinize it carefully to ensure accuracy.
For this job, you need excellent computer knowledge and organizational skills. You need the ability to pay attention to details and ability to meet deadlines. You also need communication skills and self motivation.
Strong computer skills are necessary since you will work to review materials using your computer. Organizational skills are necessary for establishing correct classifications of objects. Attention to details is crucial since you need to note any possible distinctions between the objects within an image.
As with other online jobs like microtasking, the exact skills that an employer will require from an applicant will vary according to the job. Look at Figure 2.
This is a job ad for 2D/3D Python, Data Annotator. They require computer literacy and attention to details for this position. You will assess quality depending on the metrics they will provide. Your skill set should include intermediate Python experience to manipulate data. You should know about photogrammetry, Photoshop or Lightroom, and Maya or other 3D software.
Every online work at home job, such as data entry and telemarketing requires a decently working computer and a suitable internet connection. Sometimes, you can read the exact specification for a job on a job post or from the FAQ page of the company you are applying to. Figure 3 shows an example.
This job ad says that the person they will hire for the 3D annotation job has to have a computer with at least 4GB RAM with (preferably) a GPU or graphics processing unit. The preferred internet speed of the company is 5MBPS, although the minimum of 3 MBPS will do. Note that these specifications are only for this specific job ad and do not apply to all image annotation jobs.
Job ads, such as in Figure 3, may also contain information on the age requirement. This example shows what most companies prefer for their applicants, which is a minimum of 18 years old. This age is considerably a legal working age in most places.
How To Apply
To apply for image annotator jobs, you may apply through the job portals where companies and recruiters list job openings such as for web search evaluator and telemarketing. You need to follow the application procedures implemented by that job portal. However, it is important that you read the information in the job ad because some employers include the application procedures you must complete if you ever wish to apply for that job.You may also use search engines like Google to look for image annotator jobs you can apply to. From the search results, you are likely to discover that there are companies offering professional services to different clients. One of them is Lionbridge, and this company also hires people to do data annotation tasks/projects, as shown in Figure 4.
This screenshot shows that if you work as data annotator for Lionbridge, your tasks will include labeling of objects in an image and labeling words in texts. You may click on “Learn More” so you could sign up for Lionbridge Smart Crowd.
Practice caution when signing up to any image annotation websites so you will not waste your time on fraud companies. One telltale sign of a scam website is asking you to pay money for signing up.
The salary of an image or data annotator varies per project and per company (among other factors).
Figure 5 shows the national average salary of the data annotators in the US, which is $17. If you work for 50 weeks a year on a 40 hours per week schedule, this means you will work for 2,000 hours in a year so from simple computations, you could earn $34,000 from data annotation.
You could keep this in mind as a reference when looking at offered salary rates for image annotation jobs. However, do not forget that salary in each job offer varies depending on various factors such as the complexity of the task. A job may only involve the simple task of bounding an image such as a man’s coat, and this could pay you the lowest possible rate. There are also jobs that may include other responsibilities, and these could pay a higher salary.
When looking at job ads, read every detail in it because it will probably include important details like work schedule. With image annotation, it is possible to enjoy flexible work schedules as you would with other online jobs like paid juror. Look at Figure 1 again. Although the job ad says you have flexible work hours, it is possible you will get morning, midday or night shifts.
Meanwhile, the job ad in Figure 3 seems to provide more flexibility since it says that you need to put in at least 20 hours of annotation work in one week. With this, you may pursue other
Benefits / Bonus/ Increase/ Training Provided
The same with payment and work schedules, other employment details, such as benefits and training, also vary. The main benefit you will get from annotation jobs is most likely the flexible working schedule.
As for the training, you have already known from the beginning of this post that the employer or company that hires you will provide the training you need to ensure you will do your job according to its guidelines.
Figures 3 and Figure 4 show that they will train you on how to do the annotation tasks.
I think it would be worthwhile to work as an image annotator. While the job might seem technical at first, it is comforting to know that companies will provide the training before you could perform the annotation tasks. You just need to read carefully about the details of the companies you are eyeing to apply to or sign up for so you will know the important information like payment and the work schedule that will fit your preference.