By clicking “Accept”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.

Key elements for a high-quality annotated dataset for Machine Learning models training

It is fundamental for us to train our community of annotators so the datasets are correctly and efficiently labeled. We provide them with clear instructions and monitor the quality of annotations performed throughout every step of the process.
Image annotation - outline of a woman walking

1. Ongoing digital training of our workforce

At Isahit we have developed the Isahit digital Academy to provide our community of contributors with a wide range of digital training. Our community of women can learn various skills to enhance their digital competencies and also professional tools.

2. Guidelines and instructions tailored for every project

High-quality of annotations implies a very good understanding of the labelling guidelines. For every project, the contributors are selected and assigned depending on their skills and their performance on previous similar projects.
We write very precise indications for our annotators according to the project specifications. Our project managers therefore record a video of an example of annotation to make it even clearer. Then they make sure the guidelines are well understood by implementing a test on the platform.

3. Internal quality guaranteed and tests

Throughout the project we run automated tests on performed tasks in order to ensure an unbiased and a unified dataset.We run an ongoing monitoring of the performance from the assigned contributors to be sure they understood the guidelines well and that the expected results are met.

4. Annotations workflow

We have built a seamless annotation process to ensure a more efficient workflow. Different levels of labelling can be ordered and customised according to each projects specificities. Suggestions and prelabelling tools are developed to enhance performance.

5. Feedback loops

Collaboration within our community of annotators is also a tool to promote efficiency and help among the annotators assigned to same project. Project managers can communicate directly with the team to ensure fastest projects delivery and good understanding of the guidelines to prevent mistakes.

6. Quantified metrics to follow productivity and quality

Real time analysis are run throughout the labeling process. It gives feedbacks to annotators and allows the project manager in charge to monitor annotators performance. Metrics are used to understand the time spent on the annotations but also to follow closely the quality of annotations. By randomly running check ups on annotated data it allows to prevent bias and to control each annotators performance.

Want to find out more
about AI as well as our Data Labeling tools and services?

Isahit has a wide range of solutions and tools that will help you train your algorithms. Click below to learn more!