The project was commissioned by a company that provides educational tools and services for remote learning.
The company's flagship project is a video studio for remote learning, recording online courses, video lessons, and conducting webinars.
A large volume of accurately labeled photos and videos is needed to train computer vision models.
The number of images needed to achieve high-quality results can vary from thousands to hundreds of thousands, depending on the task. Obtaining images is a manageable issue, but accurate labeling requires meticulous evaluation of each image and classification of necessary entities, which is labor-intensive. Precise labeling is crucial for training models, as it directly impacts the quality of the outcome.
To address this issue, the client sought the expertise of ICL Services, who have extensive experience in handling such tasks.Key Challenges
- Labeling objects and designated actions with utmost precision
Implemented our solution
The project began in early February 2023 and was executed over four weeks.
Labeling tens of thousands of images required an immense amount of human and time resources. However, the cost of hiring numerous highly paid specialists to work on the project was prohibitive for the client.
Therefore, to tackle this issue, ICL Services experts enlisted teams with extensive experience in content moderation and data labeling.This was a comprehensive approach that facilitated effective team management through seasoned managers overseeing the specialists' work and guaranteeing high-quality results in a short period.
By bringing on additional moderator teams, we managed to significantly cut down the project's budget.
Products and technologies
- CVAT (Computer Vision Annotation Tool)
Results
- A substantial decrease in implementation duration – down to 30 days – for the roll-out of the service.
- An affordable, manageable, flexible, and swiftly deployable service as the end product.
- The dataset was marked up in record time – 30,000 images and videos in just 2 weeks.