Task types in Toloka: identifying objects

We're continuing to introduce you to the types of tasks Toloka has to offer. In our last post, we looked at classification, while today we're going to focus on object identification: what it is, the kinds of tasks you might see, and who they're a good fit for.

Identifying objects — what it's all about

Toloka has tasks where you:

— Select objects in an image using a rectangle.

— Outline objects.

— Place dots in specific places, for example on parts of a face.

— Label an area in 3D — such as when you need to create a 3D box for each object of a type in the image, by adding points, and coloring it depending on the object type.

For all of the above task types, you might be asked to list the objects in the image (trees, people, cars) or fill in other information (labeling the price tag and entering the price, for instance).

• There could be from one to dozens of objects per task.

• The average completion time depends on the number of objects you need to select, the number of tasks in the set, and their complexity. When you only need to select one object, it should take less than a minute, while it might take a few minutes to select more. Completing 3D labeling takes at least half an hour.

• Payment is made after the task author checks the tasks.

• The average pay starts at $0.01 for simple tasks and $2 for 3D labeling. The total estimate depends on the task complexity and the number of images per object.

Perform basic object identification tasks

Let's look at an example.

You're given a picture featuring a variety of objects. The task is to only select the accessories. In order to do so, you need to frame each accessory and assign it a class: earrings, ring, glasses, or belt, for instance.

Accessing tasks

Here's how to get started:

1. Read through the instructions. That's where the task requester describes their requirements in detail: whether you're just selecting objects or if you need to list them, what you should bear in mind, and so on.

In the instructions for the accessory selection task we mentioned above, the requester wants you to only select the visible part of objects if they're blocked by something. If you can't tell what an object in a picture is, you don't select it.

2. Complete the training and test (if required). This helps you to understand the instructions, figure out what the requester expects from you, and, most importantly, unlock access to tasks.

3. Complete tasks properly. Here you need to remember what the requester requires and follow their instructions carefully. For example, outline areas exactly, don't skip objects in pictures even where there are a lot of them, and assign classes correctly.

Why object identification is important

Tasks like these are important for improving computer vision. By studying selected and classed areas, artificial intelligence can more accurately recognize and classify objects. Self-driving cars use labeled 3D images to better recognize road obstacles, distinguish pedestrians from inanimate objects, and respond in time.

Key aspects of object identification tasks in Toloka

1. To successfully complete these tasks, you need attention to detail, perseverance, thoroughness, precision, and good eyesight.
2. You're better off working on a computer since that will be easier, faster, and better. If you do it from your phone, it might not pass the quality check, in which case you won't get paid.
3. There are object identification tasks that are great for everyone and don't take a ton of attention. But there are also tasks that require specific knowledge (for example, anatomical labeling on X-ray images or jewelry labeling to determine metal defects) or more training (3D labeling).
4. If the task you're about to get started on has a labeling sample, you need to match it exactly. Also bear in mind the tips and hints in the instructions.
5. If your selection is off, you can always correct it.