Crowdworks
March 16, 2023
Download the file
Download
Oops! Something went wrong while submitting the form.

Data Annotation Project 01: Dog Nose-Print Identification AI model

Project: Dog nose print image collection and visibility classification


Request of client

- Take at least 5 photoshoots per dog in 7 different lighting conditions
- Provide EXIF ​​values ​​of collected images
- Collect images with high clarity of nose print (high clarity images have to be over 38%, at least)
- Judge visibility of collected images

*
If humans have fingerprints, dogs have “nose wrinkles” called “nose print.” Each nose print, like human fingerprints, has its own unique pattern.

What difficulties did the client have?

The client collected images of dog nose print internally by visiting abandoned animal shelters, dog cafes, and dog beauty academies, but it was difficult to find a place that allowed dogs to be photographed. In addition, it was hard to collect clear images because dogs were constantly moving and their noses were small. Due to delays in image collection and difficulties in obtaining high-quality data, the AI ​​model development schedule was delayed.

🧐 Reasons for choosing Crowdworks!

  • Data collection through 350K crowdworkers: Various environment/breed data can be collected within a short period of time
  • Multiple tasks can be performed simultaneously (from data collection to annotation)
  • Securing high-quality data through a methodical inspection system
  • Reasonable cost through settlement of each number of data satisfied by the client

Solution of Crowdworks

Collecting Images

The first thing client and the PM did was having a discussion about the collection conditions such as image resolution, shooting angle, shooting environment settings, and EXIF ​​information values ​​(representing various camera settings at the time of taking pictures, such as shooting time, camera model, aperture, and shutter speed).

The EXIF ​​value information of the images collected for the client's request had to be checked on the inspector's inspection screen. For this contract, the Crowdworks’s Platform team quickly supported the development, so they were able to proceed without a hitch. In addition, they had many concerns about workers giving up in the middle of the project since the nose print of a moving dog had to be photographed in various environments. With the help of the considerate PI(Project Improvement) team, an easy-to-understand guide was created for workers and inspectors, thereby reducing worker desertion.🙌🏻 

Classifying image visibility

In order to increase client data satisfaction, PM in charge had to be vigilant. After the first collection, the PM in charge requested feedback from the client. Through meticulous communication, PM was able to provide various sample images to project workers/inspectors, and accelerate the project progress. Since image visibility classification involves subjective judgment, PM agreed to proceed with a multi-assignment method in which several reviewers participate in one task. The difficulty of the inspection was expected to be high, and he has established a minimum requirement for selecting inspectors: passing the qualification test. he also received help from the PI team in organizing the questions and answers for the qualification test. PM was grateful to PI team for dedicated work on this project💞

Comments from PM in charge

There is no question mark for projects in Crowdworks. We lead projects that are considered as ‘Is this going to happen?’ to success. We have everything to satisfy clients’ needs. This project required detailed changes within the work platform used by the workers, but the Platform team quickly reflected it, and the PI team helped in issuing qualification tests to improve the quality of inspection. In addition, our teams are working hard to satisfy clients. Isn't it obvious why we have superb data quality over all the other companies? 

Other whitepapers you might like

Find out how reliable training data can give you the confidence to deploy AI

Get started