AI / ML

Fake vs real: A dataset to detect deep fakes

Tannah Cantrell
October 26, 2022

About AI Singapore:

The National Research Foundation (NRF) created Al Singapore (AISG) to establish strong national capabilities in Artificial Intelligence (AI), concentrating on four important pillars: research, technology, innovation, and maker space.

Twine provides an enterprise solution that enables the development of big customizable datasets by using their access to over 400,000 freelance operatives.

Why did AI Singapore reach out to Twine?

Al Singapore was having difficulty obtaining a dataset that was of sufficient quality and relevance for their project. Existing datasets did not give consistency on the variable they desired to work with, and they required examples that were as near to their use case as feasible.

In addition, they just did not have the resources to collect the quantity of data necessary in-house.

"We would definitely use Twine again for building any future datasets"

The Solution: 

Twine had the ability to generate a dataset that was as near to the input data as Al Singapore desired. Twine would be able to give varied demographics who would follow set guidelines in order to develop content by connecting with our network. This would provide Al Singapore with variety in location, background, and lighting, all of which would help them with their project.

Making the switch to Twine

The project required freelancers to record themselves on their phones chatting into the camera in various locales on a topic of their choosing. Prior to recording the footage, each participant had agreed to the use of their picture in the project.

The project's purpose was to create a dataset that would aid in the training and testing of machine-learning models for detecting false video material.

Benjamin Chua, an Al Engineer at Al Singapore, commented: 

"Twine's freelancers provided good quality content for our research project and the Twine team provided us with the wide variety of demographics that we required. 

Communication was excellent and we knew exactly what was happening at every stage, which made it easy for us to focus on the results. 

We would definitely use Twine again for building any future datasets". 

Conclusion

Twine aided in the creation of 750 films that covered all of the factors needed to teach Al to recognize deep fakes. In addition to properly managing the project and recruiting freelancers to generate the dataset, Twine's secure vault payment system was used to manage all payments to the freelancers.

Related Customer Stories