Overview

At Twine, we specialize in providing ethical data collection and data labeling services to train AI models catered to diverse client needs. Our recent collaboration with a leading technology company in threat detection exemplifies our expertise in creating custom video datasets for advanced algorithm training. In this case study, we showcase how we successfully addressed our client’s unique requirements to improve its threat detection capabilities.

Problem Statement

Our client approached Twine with a specific challenge—developing a face detection project using custom video datasets. Their goal was to train their threat detection algorithms using videos of actors enacting violent scenes. However, the challenge lies in capturing the right variety of scenarios and actions while maintaining a balance between realism and algorithm performance.

Solution

Understanding the client's objectives, Twine devised a comprehensive solution. We proposed creating 25 videos per group, each lasting between 30 seconds and 1 minute 30 seconds. These videos were meticulously designed to depict a range of violent actions, including fights, shoving, and strangling, performed by a diverse group of actors with varying levels of acting skills.

Implementation

To ensure the videos closely resembled real-life security camera footage, Twine employed specific filming techniques. The videos were shot in landscape orientation and filmed with static cameras mounted at a height of 4-5 meters or shot from first-floor windows. This unique perspective, coupled with 720p quality and a frame rate of 24fps, provided the authenticity required for effective algorithm training.

Results

Our Client was highly satisfied with the delivered videos and the results achieved. They expressed their happiness with the quality and effectiveness of the dataset in the initial stages of the annotation process. The collaboration allowed them to make significant progress in enhancing their threat detection algorithms giving them a competitive edge.

"We're very happy with the first batch of videos, and we'd like to move on to the next one. The results are great, and we've already begun the annotation process. Twine has exceeded our expectations, and we look forward to the next phase of our collaboration."

Conclusion

Twine's collaboration with clients highlights our ability to tailor video production to unique client requirements. By delivering high-quality custom video datasets, we have enabled clients to enhance their threat detection algorithms effectively. Our expertise in creating realistic videos while meeting specific algorithmic needs has proven instrumental in driving the success of this project.

Partner with Twine for your AI requirements, and let us help you unlock the potential of your algorithms through meticulously crafted custom datasets. Contact us or book a consultation call to discuss your project needs.

‍