Object detection — Step-by-step instructions
The challenge

We have a set of real-life photos of roads. We need to outline every traffic sign. Ultimately, we need to get a set of contours defined by an array of points that represent the road signs in each photo. In real-world tasks, annotation is usually done with a polygon shape. We chose to use a rectangular outline to simplify the task so that we can reduce costs and speed things up. Here’s what this might look like:

Video: 5 basic steps to run a project
Create
a project
Create
a task pool
Upload a file
with data
Launch
the pool
Get the
results

Create a project

Interface code

{
  "view": {
    "type": "view.list",
    "items": [
      {
        "type": "field.image-annotation",
        "image": {
          "type": "data.input",
          "path": "image"
        },
        "fullHeight": true,
        "data": {
          "type": "data.output",
          "path": "result"
        },
        "validation": {
          "type": "condition.required",
          "hint": "Please select an area"
        }
      }
    ]
  }
}

Create a task pool

Upload a file with data

Dataset
Prepare a TSV file with tasks as shown in our example.
License: CC BY 4.0

Launch the pool

Get the results

Automated review

Another way to review tasks is to ask other performers to do that. We recommend this option when you have limited resources for checking tasks yourself.

  • Create a simple binary classification project based on our demo
  • Ask the performers if the bounding boxes are correct
  • After the answers are collected, go back to the initial pool and upload review results on the Review assignments page
  • The file should contain the acceptance verdict and a comment. A template can be found in the Download results section
Application for corporate training
We are offering corporate training to help you solve existing challenges and develop
an internal team of Crowd Science Architects (CSA).
Mon Aug 02 2021 17:43:43 GMT+0300 (Moscow Standard Time)