Twine's freelancer network created a large-scale video dataset of everyday human activities (entering vehicles, exiting buildings, etc.) to help researchers develop quality algorithms for human action recognition and long-range biometric identification. This effort produced an open-source video library made available to the international research community as an alternative to YouTube datasets.

Twine assisted HyperSentience in collecting authentic audio recordings of individuals performing everyday activities in home environments. The project aimed to generate diverse, naturally-captured audio data for training context-aware AI models that enable devices like smartphones, VR/AR headsets, and smartwatches to understand user activities and adapt accordingly. The comprehensive dataset captured multiple variables including phone distance, mounting positions, location specifics, and background noise types.

Twine created custom video datasets featuring actors performing violent scenes to train and improve a client's threat detection algorithms. The project involved producing 25 videos per group (30 seconds to 1.5 minutes each) depicting fights, shoving, and strangling from security camera perspectives. Videos were delivered in 720p quality at 24fps with static camera angles positioned 4-5 meters high.

Twine collaborated with a global leader in human behavior analysis using face and voice sensing AI to develop a video dataset for training facial recognition and voice analysis algorithms. The initiative supported clinicians and patients in assessing mental health and emotional states through advanced AI technology. Data collection occurred monthly over approximately one year with diverse participants including varying facial shapes, sizes, skin tones, and ages.

Twine assisted HyperSentience in collecting authentic audio recordings of individuals performing everyday activities in home environments. The project aimed to generate diverse, naturally-captured audio data for training context-aware AI models that enable devices like smartphones, VR/AR headsets, and smartwatches to understand user activities and adapt accordingly. The comprehensive dataset captured multiple variables including phone distance, mounting positions, location specifics, and background noise types.

Twine created custom video datasets featuring actors performing violent scenes to train and improve a client's threat detection algorithms. The project involved producing 25 videos per group (30 seconds to 1.5 minutes each) depicting fights, shoving, and strangling from security camera perspectives. Videos were delivered in 720p quality at 24fps with static camera angles positioned 4-5 meters high.